This page intentionally left blank


without the written permission of the publisher. electronic or mechanical. ISBN : 978-81-224-2432-4 PUBLISHING FOR ONE WORLD NEW AGE INTERNATIONAL (P) LIMITED. New Delhi .110002 Visit us at www. microfilm. Ansari Road. Publishers Published by New Age International (P) . or any other means. No part of this ebook may be reproduced in any form. Daryaganj. Publishers All rights reserved. All inquiries should be emailed to rights@newagepublishers. by photostat. or incorporated into any information retrieval system..Copyright © 2007 New Age International (P) Ltd..newagepublishers. PUBLISHERS 4835/24.

. .Dedicated to our beloved Grandparents..

This page intentionally left blank .

Chapter 2 covers the types of knowledge and learning concepts. Each and every chapter has exercise. has been put into separate chapters so that courses in a variety of levels can be taught from this book. tuning of data warehouse. Chapter 8 deals with partitioning strategy in software and hardware. A brief synopsis of the chapters and comments on their appropriateness in a basic course is therefore appropriate. Chapter 3 covers how to discover the knowledge hidden in the databases using various techniques. commercial organizations.PREFACE This book is intended as a text in data mining and warehousing for engineering and post-graduate level students. etc. which are essential to data mining techniques. This book also contains sufficient material to make up an advanced course on data mining and warehousing. . Chapter 5 covers various applications related to data mining techniques and real time usage of these techniques in various fields like business. Advanced material. Chapter 4 deals with various types of data mining techniques and algorithms available. which deals both data mining techniques and data warehousing concepts suited for the modern time. Chapter 7 covers how to design the data warehouse using fact tables. dimension tables and schemas. We have attempted to cover the major topics in data mining and warehousing in depth. Chapter 6 deals with the evaluation of data warehouse and need of the new design process for databases. this book is the only book. benefits and new architecture of the data warehouse. Chapter 1 introduces the basic concepts to data mining and warehousing and its essentials. Chapter 9 covers the data mart and meta data. Chapter 11 deals with performance. We conclude that. challenges. Chapter 10 covers backup and recovery process. Model question paper is also given. however.

This page intentionally left blank .

We are also indebted to our publishers Messrs.. S. each of us wishes to “doff his hat” to the other for the unstinting efforts to make this book the best possible.ACKNOWLEDGEMENT Thanks and gratitude is owed to many kind individuals for their help in various ways in the completion of this book.L. We particularly thank my graduate students for contributing a great deal to some of the original wording and editing. and for the faith in our joint effort. Venkatesan . Retd. We first of all record our gratitude to the authorities of Bharathiyar College of Engineering and Technology. Rajagopalan. for sparing his precious time in editing this book.A. We owe thanks to our colleagues and friends who have added to this book by providing us with their valuable suggestions. for their tolerance of our many requests and especially for their commitment to this book and its authors. N. S. Without such help.R. Once again we acknowledge our family members and wish them to know that our thanks are not merely “Proforma”. Scientist. Such excellence as the book may have is owed in part to them. they are sincerely offered and well deserved. Ph. We began writing this book an year ago as good friends and we completed it as even better friends. any errors are ours. whose efforts maintained the progress of our work and ultimately put us in the right path. And finally. Bangalore. Karaikal. for the willingness to accept the advice of the other.D. Many experts made their contributions to the quality of writing by providing oversight and critiques in their special areas of interest. New Age International (P) Ltd. this book would indeed be poorer. We particularly thank Dr. Prabhu N.

This page intentionally left blank .

1 Introduction 4.3 Stages of the Data Mining Process 3.5 1.4 1.CONTENTS Preface Acknowledgement Chapter 1 Data 1.3 1.3 Anatomy of Data Mining 2.1 1.1 Introduction 2.2 Classification vii ix 1-7 1 2 3 3 4 6 7 8-13 8 8 9 12 13 14-22 14 15 15 20 20 22 23-53 23 23 Chapter 2 Chapter 3 Chapter 4 .2 Evaluation of Data Mining 3.4 Different Types of Knowledge Exercise Knowledge Discovery Process 3.1 Introduction 3.6 Mining and Warehousing Concepts Introduction Data Mining Definitions Data Mining Tools Applications of Data Mining Data Warehousing and Characteristics Data Warehouse Architecture Exercise Learning and Types of Knowledge 2.2 What is Learning? 2.2 1.4 Data Mining Operations 3.5 Architecture of Data Mining Exercise Data Mining Techniques 4.

3 Data Mining Products Exercise Data Warehouse Evaluation 6.4 4.4 OLAP and OLAP Server 6.3 Fundamental of Database 6.1 Introduction 8.4 Goals of Data Warehouse Architecture 7.2 Hardware Partitioning 8.5 Data Warehouse Users 7.3 4.2 Future Scope 5.2 The Central Data Warehouse 7.7 4.3. Information and Knowledge 6.3 RAID Levels 8. OLAP and Data Mining Exercise Data Warehouse Design 7.(xii) 4.1 Applications of Data Mining 5.7 Data Warehousing Schemas Exercise Partitioning in Data Warehouse 8.4 Software Partitioning Methods Exercise 23 24 32 34 45 46 51 52 53 54-62 54 59 61 62 63-71 63 65 65 68 69 71 72-85 72 72 74 77 78 79 81 85 86-94 86 86 88 92 94 .10 Chapter 5 Chapter 6 Chapter 7 Chapter 8 Neural Networks Decision Trees Genetic Algorithm Clustering Online Analytic Processing (OLAP) Association Rules Emerging Trends in Data Mining Data Mining Research Projects Exercise Real Time Applications and Future Scope 5. Data Warehousing Objects 7.2 Data.6 Design the Relational Database and OLAP Cubes 7. OLTP.1 Introduction 7.8 4.5 Data Warehouses.5 4.1 The Calculations for Memory Capacity 6.6 4.9 4.

3 9.6 New Architecture of Data Warehouse Exercise Glossary Multiple-Choice Questions Frequently Asked Questions & Answers Model Question Papers Bibliography Index 95-100 95 95 96 100 100 101-105 101 101 102 104 105 106-109 106 106 107 108 108 109 109 110-114 115-116 117-119 120-124 125-126 127-129 .2 Types of Backup 10.2 9.3 Challenges of the Data Warehouse 11.4 Chapter 10 Chapter 11 Appendix Appendix Appendix Appendix A B C D Mart and Meta Data Introduction Data Mart Meta Data Legacy Systems Exercise Backup and Recovery of the Data Warehouse 10.5 Future of the Data Warehouse 11.2 Prioritized Tuning Steps 11.4 Benefits of Data Warehousing 11.4 Data Warehouse Recovery Models Exercise Performance Tuning and Future of Data Warehouse 11.1 Introduction 10.1 9.(xiii) Chapter 9 Data 9.1 Introduction 11.3 Backup the Data Warehouse 10.

This page intentionally left blank .

. medical history data in health care. There are many examples that can be cited. There was also the introduction of new machine learning methods for knowledge representation based on logic programming etc. The data storage bits/bytes are calculated as follows: ² 1 byte = 8 bits ² 1 kilobyte (K/KB) = 2 ^ 10 bytes = 1. Point of sale data in retail. made data cheap. the cost of processing power and storage is falling.1 The growing base of data Data storage became easier as the availability of large amounts of computing power at low cost i. It has been estimated that the amount of information in the world doubles every 20 months and the sizes as well as number of databases are increasing even faster.e. in addition to traditional statistical analysis of data. financial data in banking and securities.1 INTRODUCTION The past couple of decades have seen a dramatic increase in the amount of information or data being stored in electronic format.024 bytes . Volume of data 1970 1980 1990 2000 Year Fig. 1. This accumulation of data has taken place at an explosive rate. policy and claim data in insurance.CHAPTER 1 DATA MINING AND WAREHOUSING CONCEPTS 1. are some instances of the types of data that is being collected. The new methods tend to be computationally intensive hence a demand for more processing power.

are good at putting data into databases quickly. Database Management Systems gave access to the data stored but this was only a small part of what could be gained from the data. According to Marcel Holshemier and Arno Siebes “Data mining is the search for relationships and global patterns that exist in large databases but are ‘hidden’ among the vast amount of data. Data Mining. safely and efficiently but are not good at delivering meaningful analysis in return. or Knowledge Discovery in Databases (KDD) as it is also known. According to William J. This is where Data Mining or Knowledge Discovery in Databases (KDD) has obvious benefits for any enterprise.125. such as clustering. and potentially useful information from data. also called as data archeology.2 | Data Mining and Warehousing ² 1 megabyte (M/MB) = 2 ^ 20 bytes = 1.921. The data is often voluminous. It is the computer which is responsible for finding .576 bytes ² 1 gigabyte (G/GB) = 2 ^ 30 bytes = 1.2 DATA MINING DEFINITIONS The term data mining has been stretched beyond its limits to apply to any form of data analysis.899. Frawley.624 bytes ² 1 exabyte (E/EB) = 2 ^ 60 bytes = 1. prediction. Gregory Piatetsky-Shapiro and Christopher J. Some of the numerous definitions of Data Mining. Matheus ‘Data Mining.099. This encompasses a number of different technical approaches. is the nontrivial extraction of implicit.606. it is the hidden information in the data that is useful” Data mining is concerned with the analysis of data and the use of software techniques for finding patterns and regularities in sets of data. analyzing changes. finding dependency networks. These relationships represent valuable knowledge about the database and the objects in the database and.152. forecasting and estimation.776 bytes ² 1 petabyte (P/PB) = 2 ^ 50 bytes = 1. data summarization.741. is the process of extracting hidden knowledge from large volumes of raw data and using it to make crucial business decisions.048.976 bytes ² 1 zettabyte (Z/ZB) =1 000 000 000 000 000 000 000 bytes ² 1 yottabyte (Y/YB) =1 000 000 000 000 000 000 000 000 bytes It was recognized that information is at the heart of business operations and that decisionmakers could make use of the data stored to gain valuable insight into the business. data dredging.073.846. Traditional on-line transaction processing systems. Data mining refers to “using a variety of techniques to identify nuggets of information or decision-making knowledge in bodies of data. Analyzing data can provide further knowledge about a business by going beyond the data explicitly stored to derive knowledge about the business. data harvesting.504.842. but as it stands of low value as no direct use can be made of it. and extracting these in such a way that they can be put to use in the areas such as decision support. and detecting anomalies’. OLTPs.627. 1. previously unknown.906. such as a relationship between patient data and their medical diagnosis. if the database is a faithful mirror.824 bytes ² 1 terabyte (T/TB) = 2 ^ 40 bytes = 1. or Knowledge Discovery in Databases are: Extraction of interesting information or patterns from data in large databases is known as data mining. of the real world registered by the database”. learning classification rules.511.

making use of as much of the collected data as possible to arrive at reliable conclusions and decisions. Data mining analysis tends to work from the data up and the best techniques are those developed with an orientation towards large volumes of data. 1. Once knowledge has been acquired this can be extended to larger sets of data working on the assumption that the larger data set has a structure similar to the sample data. ORACLE9i. 1. The analysis process starts with a set of data. By delivering complete programmatic control of the database in data mining. scalable modeling and real-time scoring. . The idea is that it is possible to strike gold in unexpected places as the data mining software extracts patterns not previously discernable or so obvious that no-one has noticed them before.3 DATA MINING TOOLS The best of the best commercial database packages are now available for data mining and warehousing including IBM DB2. 4 Thought and SYBASE System 10. Clementine.1 Oracle Data Mining (ODM) Oracle enables data mining inside the database for performance and scalability. This integrated intelligence enables the automation and decision speed that e-businesses require in order to compete today. Again this is analogous to a mining operation where large amounts of low-grade materials are sifted through in order to find something of value.Data Mining and Warehousing Concepts | 3 the patterns by identifying the underlying rules and features in the data. Oracle Data Mining (ODM) delivers powerful. INFORMIX-On Line XPS.3. delivering accurate insights completely integrated into e-business applications. This enables e-businesses to incorporate predictions and classifications in all processes and decision points throughout the business cycle. Some of the capabilities are: ² An API that provides programmatic control and application integration ² Analytical capabilities with OLAP and statistical functions in the database ² Multiple Algorithms: Naïve Bayes and Association Rules ² Real-time and Batch Scoring modes ² Multiple Prediction types ² Association insights ORACLE9i Data Mining provides a Java API to exploit the data mining functionality that is embedded within the ORACLE9i database.4 APPLICATIONS OF DATA MINING Data mining has many and varied fields of application some of which are listed below. uses a methodology to develop an optimal representation of the structure of the data during which time knowledge is acquired. 1. Intelligent Miner. ODM is designed to meet the challenges of vast amounts of data.

2 Banking ² ² ² ² ² ² Credit card fraudulent detection Identify ‘loyal’ customers Predict customers likely to change their credit card affiliation Determine credit card spending by customer groups Find hidden correlation’s between different financial indicators Identify stock trading rules from historical market data 1. As well as integrating data throughout an enterprise.1 Sales/Marketing ² ² ² ² Identify buying patterns from customers Find associations among customer demographic characteristics Predict response to mailing campaigns Market basket analysis 1. which medical procedures are claimed together Predict which customers will buy new policies Identify behaviour patterns of risky customers Identify fraudulent behaviour 1. A data warehouse is a relational database management system (RDBMS) designed specifically to meet the needs of transaction processing systems.e.5 Medicine ² Characterize patient behaviour to predict office visits ² Identify successful medical therapies for different illnesses 1.4 | Data Mining and Warehousing 1. Data mining potential can be enhanced if the appropriate data has been collected and stored in a data warehouse. analyst) to make better and faster decisions.4. manager.4 Transportation ² Determine the distribution schedules among outlets ² Analyze loading patterns 1.4. It can be loosely defined as any centralized data repository which can be queried for business benefit but this will be more clearly defined later..4. format. Data warehousing is a new powerful technique making it possible to extract archived operational data and overcome inconsistencies between different legacy data formats.3 Insurance and Health Care ² ² ² ² Claims analysis i.4. or communication requirements it is possible to incorporate additional or expert information. regardless of location. . aimed at enabling the knowledge worker (executive.4.5 DATA WAREHOUSING AND CHARACTERISTICS Data warehousing is a collection of decision support technologies.

and loading (ETL) solution. There are a number of ETL tools available in the market to do ETL process the data according to business/technical requirements. client analysis tools. Data warehouses must put data from disparate sources into a consistent format. Integration is closely related to subject orientation. you can answer questions like “Who was our best customer for this item last year?” This ability to define a data warehouse by subject matter. Using this warehouse. These difficulties are eliminated by ETL Tools since they are very powerful and they offer many advantages in all stages of ETL process starting from extraction. the above mentioned ETL process was done manually by using SQL code created by programmers. When they achieve this. transform and load the data into Data Warehouse for decision making. This task was tedious in many cases since it involved many resources. transformation. is as follows: ² Subject Oriented ² Integrated ² Nonvolatile ² Time Variant Data warehouses are designed to help you analyze data. they are said to be integrated. makes the data warehouse subject oriented. ETL Tools are meant to extract. On top of it. an online analytical processing (OLAP) engine. . you can build a warehouse that concentrates on sales. Following are some of those: Tool name Informatica DT/Studio Data Stage Ab Initio Data Junction Oracle Warehouse Builder Microsoft SQL Server Integration Transform On Demand Solonde Transformation Manager Company name Informatica Corporation Embarcadero Technologies IBM Ab Initio Software Corporation Pervasive Software Oracle Corporation Microsoft ETL Solutions A common way of introducing data warehousing is to refer to the characteristics of a data warehouse as set forth by William Inmon. data profiling. data cleansing. a data warehouse environment includes an extraction. For example. debugging and loading into data warehouse when compared to the old method. and other applications that manage the process of gathering data and delivering it to business users. author of Building the Data Warehouse and the guru who is widely considered to be the originator of the data warehousing concept. to learn more about your company’s sales data. transportation. complex coding and more work hours. Before the evolution of ETL Tools. maintaining the code placed a great challenge among the programmers. They must resolve such problems as naming conflicts and inconsistencies among units of measure.Data Mining and Warehousing Concepts | 5 In addition to a relational database. sales in this case. transformation.

g. marketing. transforming and integrating this data. to be used for comparisons. A data warehouse’s focus on change over time is what is meant by the term time variant. or finance. 1. In a dependent data mart. data can be collected directly from sources. gender data is transformed to “m” and “f”. Data in the warehouse and data marts is stored and managed by Monitoring & Administration Metadata Repository OLAP Servers Data Warehouse External Sources Extract Transform Load Refresh Query/Reporting Serve Data Mining Analysis Operational dbs Data Sources Data Marts Fig. trends. perhaps onto slower archival storage. the data can be derived from an enterprise-wide data warehouse. In addition to the main warehouse. Nonvolatile means that. once entered into the warehouse. in one application. This is logical because the purpose of a warehouse is to enable you to analyze what has occurred. This is very much in contrast to online transaction processing (OLTP) systems. analysts need large amounts of data. When data are moved from the operational environment into the data warehouse. gender might be coded as “m” and “f” in another by 0 and 1.6 | Data Mining and Warehousing For instance. In an independent data mart. or older. The data warehouse contains a place for storing data that are 10 to 20 years old. and for periodically refreshing the warehouse to reflect updates at the sources and to purge data from the warehouse. A data warehouse that is designed for a particular line of business. for loading data into the data warehouse. data should not change.6 DATA WAREHOUSE ARCHITECTURE Data warehouse architecture includes tools for extracting data from multiple operational databases and external sources. 1. and forecasting.2 Data warehousing architecture Tools . such as sales. they assume a consistent coding convention e. there may be several departmental data marts. These data are not updated. where performance requirements demand that historical data be moved to an archive. In order to discover trends in business. for cleaning.

Explain the uses of data mining and describe any three data mining softwares. . 2. report writers. Finally. and data mining tools. 4. EXERCISE 1. Explain the data warehouse architecture. 5. and tools for monitoring and administering the warehousing system. which present multidimensional views of data to a variety of front end tools: query tools. What is data warehouse? List out the characteristics of data warehouse. there is a repository for storing and managing meta data.Data Mining and Warehousing Concepts | 7 one or more warehouse servers. 3. analysis tools. Define data mining.

Generally it is only possible to use a small number of properties to characterize objects so we make abstractions in that objects which satisfy the same subset of properties are mapped to the same internal representation. The nature of the environment is dynamic hence the model must be adaptive i. living creatures must have the ability to adapt themselves to their environment. Inductive learning where the system infers knowledge itself from observing its environment has two main strategies: ² Supervised learning—This is learning from examples where a teacher helps the system construct a model by defining classes and supplying examples of each class. There are undeniably many different ways of learning but instead of considering all kinds of different definitions. should be able to learn. database is analyzed with a view to finding patterns.e. the common properties in the examples.1 Inductive learning Induction is the inference of information from data and inductive learning is the model building process where the environment i. Once ..CHAPTER 2 LEARNING AND TYPES OF KNOWLEDGE 2. 2.. we need two other concepts a certain ‘task’ to be carried out either well or badly and a learning ‘subject’ that is to carry out the task. Similar objects are grouped in classes and rules formulated whereby it is possible to predict the class of unseen objects. In order to define learning operationally. This process of classification identifies classes such that each class has a unique pattern of values which forms the class description. 2. we will not state what learning is but rather specify how to determine when someone has learned something.1 INTRODUCTION It seems that learning is one of the basic necessities of life.e. we will concentrate on an operational definition of the concept of learning.e.2. The system has to find a description of each class i..2 WHAT IS LEARNING? An individual learns how to carry out a certain task by making a transition from a situation in which the task cannot be carried out to a situation in which the same task can be carried out under the same circumstances.

. changes within. i. This is similar to discriminate analysis as in statistics. policy and claim data in insurance. The data mine system is supplied with objects but no classes are defined so it has to observe the examples and recognize patterns (i. Point of sale data in retail. which states that if there are multiple explanations for a particular phenomenon it makes sense to choose the simplest because it is more likely to capture the nature of the phenomenon. financial data in banking and securities.Learning and Types of Knowledge | 9 the description has been formulated the description and the class form a classification rule which can be used to predict the class of previously unseen objects..e. and the fields in the record correspond to attributes. The simpler models are more likely to be correct if we adhere to Ockham’s razor. This system results in a set of class descriptions. ² Unsupervised learning—This is learning from observation and discovery. There are many examples that can be cited.g..e. 2. Database Systems Statistics Machine Learning Data Mining Visualization Algorithm Other Disciplines Fig. The problem is that most environments have different states. with some being numerical (continuous valued. 2.1 Multi-disciplinary approach in data mining . are some instances of the types of data that is being collected. disease). and it is not always possible to verify a model by checking it for all possible situations. one for each class discovered in the environment.. Again this similar to cluster analysis as in statistics. The quality of the model produced by inductive learning methods is such that the model could be used to predict the outcome of future situations in other words not only for states encountered but rather for unseen states that could occur. medical history data in health care.3 ANATOMY OF DATA MINING The use of computer technology in decision support is now widespread and pervasive across a wide range of business and industry. where each individual record may correspond to a transaction or a customer. Induction is therefore the extraction of patterns. e. class description) by itself. Given a set of examples the system can construct multiple models some of which will be simpler than others. This has resulted in the capture and availability of data in immense volume and proportion. age) and some symbolic (discrete valued.g. these fields are of mixed type. e. The data are typically a collection of records. Very often.

learning with teacher. This set contains examples i. real-world databases. Machine learning examines previous examples and their outcomes and learns how to reproduce these and make generalizations about new cases. Data mining however allows the expert’s knowledge of the data and the advanced analysis techniques of the computer to work together. but not of KDD. there are efforts to extract knowledge from neural networks which are very relevant for KDD.2. ² KDD is concerned with very large. observations coded in some machine readable form. When integrating machine-learning techniques into database systems to implement KDD some of the databases require: . However.3. Generally a machine learning system does not use single observations of its environment but an entire finite set called the training set at once. 2. So efficiency questions are much more important for KDD.e. Analysts to detect unusual patterns and explain patterns using statistical models such as linear models have used statistical analysis systems such as SAS and SPSS. etc. ² ML is a broader field which includes not only learning from examples. Statistics have a role to play and data mining will not replace such analyses but rather they can act upon more directed analyses based on the results of data mining.1 Differences between Data Mining and Machine Learning Knowledge Discovery in Databases (KDD) or Data Mining.2 Machine Learning Machine learning is the automation of a learning process and learning is tantamount to the construction of rules based on observations of environmental states and transitions. a concept representing the results of learning as output. while ML typically (but not always) looks at smaller data sets. but also reinforcement learning.10 | Data Mining and Warehousing 2.3.3. learning with teacher. 2.1 Statistics Statistics has a solid theoretical foundation but the results from statistics can be overwhelming and difficult to interpret as they require user guidance as to where and how to analyze the data. The main differences are: ² KDD is concerned with finding understandable knowledge. while ML is concerned with improving performance of an agent. The training set is finite hence not all concepts can be learned exactly. For example statistical induction is something like the average rate of failure of machines. etc.. So training a neural network to balance a pole is part of ML. A learning algorithm takes the data set and its accompanying information as input and returns a statement e. KDD is that part of ML which is concerned with finding understandable knowledge in large sets of real-world examples. and the part of Machine Learning (ML) dealing with learning from examples overlap in the algorithms used and the problems addressed..g. but also reinforcement learning. This is a broad field which includes not only learning from examples.

//Data spaces being the number of examples.3 Database Systems A database is a collection of information related to a particular subject or purpose.The presentation ranges from simple table to complex graph. rules in a rule-based system. and ² Interpreting the knowledge produced to solve users’ problems and/or reduce data spaces. ² More expressive representations for both data. which represent instances of a problem domain.2 Statistical Algorithms Statistics is the science of colleting.3.4 Algorithms 2.. tuples in relational databases.g. e. Practical KDD systems are expected to include three interconnected phases ² Translation of standard database information into a form suitable for use by learning facilities.3. and the semantic information contained in the relational schemata.. which can be used to solve users’ problems in the domain. ² Using machine learning techniques to produce knowledge bases from databases. 2.g. e. Data mining allows the analyst to focus on certain patterns and trends and explore in-depth using visualization. and knowledge. more intuitive understanding of the data and as such can work well along side data mining. 2. It is usual that the database is often designed for purposes different from data mining and so properties or attributes that would simplify the learning task are not present nor can they be requested from the real world. Visualization indicates the wide range of tool for presenting the reports to the end user . and natural selection in a design based on the concepts of natural evolution. organizing.1 Genetic Algorithms Optimization techniques that use processes such as genetic combination.Learning and Types of Knowledge | 11 ² More efficient learning algorithms because realistic databases are normally very large and noisy..4.3. and applying numerical facts. Statistical analysis systems such as SAS and SPSS have been used by analysts to detect unusual patterns and explain patterns using statistical models such as linear models. Databases are usually contaminated by errors so the data mining algorithm has to cope with noise whereas ML has laboratory type examples i.3. such as tracking customer orders or maintaining a music collection.5 Visualization Data visualization makes it possible for the analyst to gain a deeper. 2. mutation. as near perfect as possible.4. .e. 2. using various 2D and 3D rendering techniques to distinguish information presented. On its own data visualization can be overwhelmed by the volume of data in a database but in conjunction with data mining can help with exploration.3.

It is almost impossible to decipher a message that is encrypted if you do not have key. 2.3 Hidden Knowledge This is data that can be found relative easily by using pattern recognition or machine learning algorithms. 2. one could use SQL to find these patterns but this would probably prove extremely time-consuming. Deep knowledge is typically the result of a search space over only a tiny local optimum.4. hilly landscape.4.4. 2.4 DIFFERENT TYPES OF KNOWLEDGE Knowledge is a collection of interesting and useful pattern in a database. for the present at any rate. An example of this is encrypted information stored in a database. A search algorithm could roam around this landscape for ever. there is a limit to what one can learn.1 Shallow Knowledge This is information that can be easily retrieved from database using a query tool such as Structured Query Language (SQL). However. which indicates that. whereas you would have to spend months using SQL to achieve the same result. The advantage of OLAP tools is that they are optimized for the kind of search and analysis operation. without achieving any significant result.12 | Data Mining and Warehousing 2. with no indication of any elevations in the neighborhood. Again. it cannot search for optimal solutions. Here information that can be obtained through data mining techniques. OLAP is not as powerful as data mining. a search algorithm can easily find a reasonably optimal solution. TABLE 2. With and different orderings of the data but it is important to realize that most of the things you can do with an OLAP tool can also be done using SQL.1 Different Types of Knowledge and Techniques Type of knowledge Shallow knowledge Multi-dimensional knowledge Hidden knowledge Deep knowledge Technique SQL OLAP Data mining Clues .2 Multi-Dimensional Knowledge OLAP tools you have the ability to rapidly explore all sorts of clusterings this is information that can be analyzed using online analytical processing tools. In data mining we distinguish four different types of knowledge. Hidden knowledge is the result of a search space over a gentle. 2.4.4 Deep Knowledge This is information that is stored in the database but can only be located if we have a clue that tells us where to look. A pattern recognition algorithm could find regularities in a database in minutes or at most a couple of hours. The key issue in Knowledge Discovery in Database is to realize that there is more information hidden in your data than you are table to distinguish at first sight.

Learning and Types of Knowledge | 13 EXERCISE 1. 2. What are the differences between data mining and machine learning? . 6. 3. 4. Explain in detail about self-learning computer systems and concept learning. What is learning? Explain different types of learning with example. 5. With a neat sketch explain the architecture of data mining. Write in detail about data mining and query tools. 7. What are the different kinds of knowledge? Compare data mining with query tools.

such as clustering. used to provide insight into the trends inherent in historical information. previously unknown. cleaning. Similarly. enrichment. it becomes apparent which data elements are considered more meaningful than others in terms of decision support and. Data mining and data warehouses complement each other. and transformation of data. or Knowledge Discovery in Databases (KDD) as it is also known. cleaning. Data mining does not guarantee the behavior of future data through the analysis of historical data. This encompasses a number of different technical approaches. the process of data warehousing improves as. finding dependency networks. Data mining is not a “black box” process in which the data miner simply builds a data mining model and watches as meaningful information appears. data summarization. Instead. selection. misconceptions about the role that data mining plays in decision support solutions can lead to confusion about and misuse of these tools and techniques. Data mining. data mining is a guidance tool. including research. and transformation steps that are also typically associated with data mining. in turn. analyzing changes.1 INTRODUCTION Data mining is an effective set of analysis tools and techniques used in the decision support process.CHAPTER 3 KNOWLEDGE DISCOVERY PROCESS 3. these tools work best on well-prepared data to answer well-researched business scenarios—the GIGO (garbage in. Although Analysis Services removes much of the mystery and complexity of the data mining process by providing data mining tools for creating and examining data mining models. However. Quite a bit of work. learning classification rules. Well-designed data warehouses have handled the data selection. through data mining. enrichment. improves the data cleaning and transformation steps that are so crucial to good data warehousing practices. is the nontrivial extraction of implicit. must be performed first if data mining is to truly supply meaningful information. and potentially useful information from data. garbage out) law applies more to data mining than to any other area in Analysis Services. . and detecting anomalies.

proactive information delivery.1 Data Mining Evaluation Evolutionary step Data Collection (1960s) Data Access (1980s) Business question “What was my total revenue in the last four years?” “What were unit sales in New Delhi last May?” “What were unit sales in New Delhi last March? Drill down to Chennai. Interpretation/ Knowledge 4. Prospective. Structured dynamic data delivery Query Language (SQL). Transformation 2. ODBC On-line analytic processing (OLAP). Data Warehousing & Decision Support (1990s) Data Mining (Emerging Today) 3. Data Prep 1. static data delivery. 3.2 EVALUATION OF DATA MINING TABLE 3. disks Characteristics Retrospective. (RDBMS). data warehouses Advanced algorithms.3 STAGES OF THE DATA MINING PROCESS 5. Data Selection & Cleaning Fig. Relational databases Retrospective.Knowledge Discovery Process | 15 3. tapes.” “What’s likely to happen to Chennai unit sales next month? Why?” Enabling technologies Computers.1 An overview of the steps that compose the KDD process . multiprocessor computers. dynamic data delivery at multiple levels. at record level. massive databases Retrospective. multidimensional databases. Data Mining/ Pattern Discovery 3.

Data cleaning is separate from data enrichment and data transformation because data cleaning attempts to correct misused or incorrect attributes in existing data.16 | Data Mining and Warehousing 3. such as people making typing errors. TABLE 3. such as orphan records or required fields set to null.5 –10. tends to be more mechanical in nature than the second part. age.3. and logical inconsistencies. . In the original table we have X and X1. A important element in a cleaning operation is the de-duplication of records.5 ——— ——— 10.5 3. In our example we start with a database containing records of patient for hypertension diseases with duration. The process of inspecting data for physical inconsistencies. adds new attributes to existing data. for data mining purposes. some of which can be executed in advance while others are invoked only after pollution is detected at the coding or the discovery stage. the data is uniform in terms of key and attribute usage.2 Data Cleaning Data cleaning is the process of ensuring that. There are several types of cleaning process. Data enrichment. locating data. address. In a normal database some clients will be represented by several records.1 Data Selection There are two parts to selecting data for data mining. of the data to be examined. 101 102 103 104 105 104 Name U V W X Y XI Address UU1 VV1 WW1 XX1 YY1 XXI Sex Male Male Male Female Male Female Age 49 45 61 49 43 49 Hype_dur ——— 10. or domain. they have a lot in common and pattern recognition algorithm can be applied in cleaning data. a copy of this operational data is drawn and stored in a separate database.No. which requires significant input by a domain expert for the data. A domain expert is someone who is intimately familiar with the business purposes and aspects. while data transformation changes the form or structure of attributes in existing data to meet specific data mining requirements. disease particulars.2 Original Data P.3. or of clients moving from one place to another without notifying change of address. In order to facilitate the KDD process. designation. It is a selection operational data from the Public Health Center patients of a small village contains information about the name. Although data mining and data cleaning are two different disciplines. by contrast. such as accounts with closing dates earlier than starting dates. period of diseases. identifying data. They have same Number and Address but different name. which is a strong indication that they have the same person but that one of the spelling is incorrect. although in many cases this will be the result of negligence. The first part.

In our example we replace the unknown data to NULL. or provide additional derived attributes for a better understanding of indirect relationships.5 10. You can add information to such data from standardized external industry sources to make the data mining process more successful and reliable. to existing data. obtained from either different departments or companies or vendors that sell standardized industry-relevant data. data warehouses frequently provide pre aggregation across business lines that share common attributes for cross-selling analysis purposes. In our example. 101 102 103 104 105 Name U V W X Y Address UU1 VV1 WW1 XX1 YY1 Sex Male Male Male Female Male Age 49 45 61 49 43 Hype_dur — 10. For example. this step is best handled in a temporary storage area.Knowledge Discovery Process | 17 TABLE 3. The duration value is indicated in negative value for Patient number 103. can require a number of updates to both data and meta data.4 Domain Consistency P. we will suppose that we have purchased extra information about our patients consisting of age. As with data cleaning and data transformation. The value must be positive so the incorrect value to be replaced with NULL. such as calculated fields or data from external sources. Data transformation involves the manipulation of data.3 Data Enrichment Data enrichment is the process of adding new attributes. kidney. but data enrichment involves adding information to existing data. Most references on data mining tend to combine this step with data transformation. This type of pollution is particularly damaging. in particular the combination of external data sources with data to be mined. 101 102 103 104 105 Name U V W X Y Address UU1 VV1 WW1 XX1 YY1 Sex Male Male Male Female Male Age 49 45 61 49 43 Hype_dur NULL 10.3.5 NULL 10.3 De-duplication P.5 — Second type of pollution that frequently occurs in lack of domain consistency and disambiguation. This can include combining internal data with external data. TABLE 3. because it is hard to trace.No.5 -10. Data enrichment is an important step if you are attempting to mine marginally acceptable data.5 NULL 3. Data enrichment. heart and stroke related disease. This is more realistic than it may initially . and such updates are generally not acceptable in an established data warehouse. but in greatly influence to type of patterns when we apply data mining to this table.No.

since it is quite possible to buy demographic data on average age for a certain neighborhood and hypertension and diabetes patients can be traced fairly easily.6 Coded Database HyperTension 1 0 0 1 1 Occupation HypeDur Diabetes 0 1 1 0 0 Address Kidney Diadur Stroke 0 0 0 0 0 Name Heart 1 1 1 1 0 Age 49 45 61 49 43 Sex Male Male Male Female Male U V W X Y Labor Farmer Farmer Farmer Farmer UU1 VV1 WW1 XX1 YY1 NULL 10. A general rule states that any detection of data must be a conscious decision. For this. TABLE 3. a lot of desirable data is missing and most is impossible to retrieve. but instead grooms existing attributes for data mining purposes. is the process of changing the form or structure of existing attributes. TABLE 3. especially fraud detection. In the next stage. Data transformation is separate from data cleansing and data enrichment for data mining purposes because it does not correct existing attribute data or add new attributes. In our example we convert disease yes-no into 1 and 0. 3. in terms of data mining. In most tables that are collected from operational data.4 Data Transformation Data transformation. after a thorough analysis of the possible consequences.5 Enrichment Name U V W X Y Address UU1 VV1 WW1 XX1 YY1 Sex Male Male Male Female Male Age 49 45 61 49 43 Kidney No No No No No Heart Yes Yes Yes Yes No Stroke No No No No No 3. however. lack of information can be a valuable indication of interesting patterns. this is a situation that occurs frequently in practice. it is not particularly important how the information was gathered and be easily be joined to the existing records. we select only those records that have enough information to be of value.3. First the extra information that was purchased to enrich the database is added to the records describing the individuals.5 NULL 10.5 NULL NULL 13 14 15 NULL 0 0 0 0 0 .5 Coding That data in our example can undergo a number of transformations.3. Although it is difficult to give detailed rules for this kind of operation. In some cases.18 | Data Mining and Warehousing seen.

6 Data Mining The discovery stage of the KDD process is fascinating.3. and this clearly indicates the need for hybrid learning.3. We shall also show that there is a relationship is detected during the data mining stage. Data mining is not so much a single technique as the idea that there is more knowledge hidden in the data than shows itself on the surface. and may be used at the beginning of a data mining process to get a rough feeling of the quality of the data set and where patterns are to be found.7 Visualization/Interpretation/Evaluation Visualization techniques are a very useful method of discovering patterns in datasets. Interesting possibilities are offered by object oriented three dimensional tool kits. These simple methods can provide us with a wealth of information. Here we shall discuss some of the most important machine-learning and pattern recognition algorithms.7 Final Table after Removing Rows HyperTension 0 1 Occupation HypeDur (Years) Diabetes 1 0 Address Kidney Diadur (Years) V X Farmer Farmer VV1 XX1 Male Female 45 49 10.Knowledge Discovery Process | 19 TABLE 3. An elementary technique that can be of great value is the so called scatter diagram. Scatter diagrams can be used to identify Stroke 0 0 Name Heart Age Sex .5 10. which enable to user to explore three dimensional structures interactively. From this point of view. Advanced graphical techniques in virtual reality enable people to wander through artificial data spaces. We shall see that some learning algorithms do well on one part of the set where others fail. such as Inventor. and in this way get an idea of the opportunities that are available as well as some of the problems that occur during the discovery stage. so data mining techniques from quite a heterogeneous group. those that are of interest in present context are: ² Query tools ² Statistical techniques ² Visualization ² Online analytical processing ² Case based learning (k-nearest neighbor) ² Decision trees ² Association rules ² Neural networks ² Genetic algorithms 3.5 13 15 0 0 1 1 3. white historic development of data sets can be displayed as a kind of animated movie. data mining is really an ‘anything goes’ affair. Although various different techniques are used for different purposes. Any technique that helps extract more out of our data in useful.

3. or due to causal reasons.1 Creation of Prediction and Classification Models This is the most commonly used operation primarily because of the proliferation of automatic model development techniques.e.4. 3.4.20 | Data Mining and Warehousing interesting subsets of the data sets so that we can focus on the rest of the data mining process. i.2 Association Whereas the goal of the modeling operation is to create a generalized description that characterizes the contents of a database. since many data mining modeling techniques express models as sets of if.. its goal is to identify outlying points in a particular data set. Model creation has been traditionally pursued using statistical techniques. Database.4. .4 DATA MINING OPERATIONS Four operations are associated with discovery-driven data mining. and explain whether they are due to noise or other impurities being present in the data. to automatically generate a model that can predict a future behavior. 3.3 Database Segmentation As databases grow and are populated with diverse types of data it is often necessary to partition them into collections of related records either as a means of obtaining a summary of each database. There is a whole field of research dedicated to the search for interesting projections for data setsthat is called projection pursuit. 3. It is usually applied in conjunction with database segmentation. data warehouses. Data cleaning and data integration techniques may be performed on the data.4 Deviation Detection This operation is the exact opposite of database segmentation.. rules. data warehouse of a typical data mining system may have the following components.4. the goal of Association is to establish relations between the records in a data base. In particular. It is usually the source of true discovery since outliers express deviation from some previously known expectation and norm.. The value added by data mining techniques in this operation is in their ability to generate models that are comprehensible. data about the past. spreadsheets or other kinds of information repositories.. The goal of this operation is to use the contents of the database. 3.5 ARCHITECTURE OF DATA MINING Based on databases. which reflect historical data. data warehouse or other information repository This is one or a set of databases. 3. or before performing a data mining operation such as model creation. then. and explainable..

Graphical Interface Pattern Evaluation Data Mining Engine Knowledge Base Database or Data Warehouse Server Data Cleaning Data Integration Database Filtering Data Warehouse Fig.Knowledge Discovery Process | 21 Database or data warehouse server The database or data warehouse server is responsible for fetching the relevant data.2 Architecture of data mining Knowledge base This is the domain knowledge that is used to guide the search or evaluate the interestingness of resulting patterns. Data mining engine This is essential to the data mining system and ideally consists of a set of functional modules for tasks such as characterization. classification. 3. . based on the user’s data mining request. may also be included. association. which can be used to assess a pattern’s interestingness based on its unexpectedness. Such knowledge can include concept hierarchies. cluster analysis and evolution and deviation analysis. used to organize attributes or attribute values into different levels of abstraction. Knowledge such as user beliefs.

this component allows the user to browse database and data warehouse schemes or data structures. 4. allowing the user interact with the system by specifying a data mining query or task. Explain four operations that are associated with discovery-driven data mining.22 | Data Mining and Warehousing Pattern evaluation module This component typically employs interestingness measures and interacts with the data mining modules so as to focus the search towards interesting patterns. For efficient data mining. With a neat sketch. Write a short note on evaluation of data mining. explain the architecture of a data mining system. EXERCISE 1. (ii) Data cleaning. 3. Explain the various steps involved in Knowledge Discovery Process. Graphical user interface This module communicates between user and the data mining system. (iii) Data enrichment. In addition. based on the intermediate data mining results. Write short notes on (i) Data selection. evaluate mined patterns and visualize the patterns in different forms. 2. it is highly recommended to push the evaluation of pattern interestingness as deep as possible into the mining process so as to confine the search to only the interesting patterns. . 5. providing information to help focus the search and performing exploratory data mining.

1 INTRODUCTION There are several different methods used to perform data mining tasks.3 NEURAL NETWORKS Neural Networks are analytic techniques modeled after the (hypothesized) processes of learning in the cognitive system and the neurological functions of the brain and capable of predicting new NEURONS INPUT LAYER 1 LAYER 2 OUTPUT Fig. 4. These techniques not only require specific types of data structures. In this chapter. but also imply certain types of algorithmic approaches. Prediction involves using some variables or fields in the database to predict unknown or future values of other variables of interest. we briefly examined some of the data mining techniques. 4. Examples of classification methods used as part of knowledge discovery applications include classifying trends in financial markets and automated identification of objects of interest in large image databases.1 Neural networks .CHAPTER 4 DATA MINING TECHNIQUES 4. Description focuses on finding human interpretable patterns describing the data.2 CLASSIFICATION Classification is learning a function that maps a data item into one of several predefined classes. 4.

create a ‘N’ node with T as its parent and stop.1 Constructing Decision Trees Decision tree programs construct a decision tree T from a set of training cases. which variables matter.” In that phase.4. neurons apply an iterative process to the number of inputs (variables) to adjust the weights of the network in order to optimally predict (in traditional terms one could say. 2. 3.24 | Data Mining and Warehousing observations (on specific variables) from other observations (on the same or other variables) after executing a process of so-called learning from existing data. Neural Networks techniques can also be used as a component of analyses designed to build explanatory models because Neural Networks can help explore data sets in search for relevant variables or groups of variables. The algorithm consists of five steps. Decision trees represent rules. create a ‘P’ node with T as its parent and stop. Decision tree is a classifier in the form of a tree structure where each node is either: ² a leaf node. If all examples in T are positive. . or even to some extent. It is virtually impossible to “interpret” the solution in traditional. or ² a decision node that specifies some test to be carried out on a single attribute value. 4. 4. After the phase of learning from an existing data set. 1. Create a T node. which provides the classification of the instance. analytic terms. with one branch and sub-tree for each possible outcome of the test. theoretically. The original idea of construction of decision trees goes back to the work of Hoveland and Hunt on Concept Learning Systems (CLS) in the late 1950s. the neural network is ready and it can then be used to generate predictions. The neural network is then subjected to the process of “training. T ← the whole training set.4 DECISION TREES Decision trees are powerful and popular tools for classification and prediction. Disadvantages The final solution depends on the initial conditions of the network. Advantages Neural Networks is that. they are capable of approximating any continuous function. the results of such explorations can then facilitate the process of model building. and thus the researcher does not need to have any hypotheses about the underlying model. A decision tree can be used to classify an instance by starting at the root of the tree and moving through it until a leaf node. indicating a class of instances. find a “fit” to) the sample data on which the “training” is performed. If all examples in T are negative. such as those used to build theories that explain phenomena.

4. 4. Example for decision tree . The key requirements to do mining with decision trees are: Predefined classes: The categories to which cases are to be assigned must have been established beforehand (supervised data). …. no. .. values on X.1 B < 4.5 K=x K=y K=x C = true K=x C = false K=y C A = blue B B < 8.. Machine Learning. vN and partition T into subsets T1. Sufficient data: Usually hundreds or even thousands of training cases.leaf node Fig. Select an attribute TN according their and X = vi as the 5..Data Mining Techniques | 25 4. T2. and there must be for more cases than classes. Ross Quinlan originally developed ID3 at the University of Sydney.decision node A A = red B B > 8.. ← Ti and go to step 2.4.1 . vol.2 ID3 Decision Tree Algorithm J. For each Ti do: T X with values v1. N) with T as their parent label of the branch from T to Ti. 1. Discrete classes: A case does or does not belong to a particular class.5 B > 4. 1. He first presented ID3 in 1975 in a book. …. v2. “Logical” classification model: Classifier that can be only expressed as decision trees or set of production rules. ID3 is based on the Concept Learning System (CLS) algorithm. Create N nodes Ti (i = 1.2 An example of a simple decision tree Decision tree induction is a typical inductive approach to learn knowledge on classification.

if pp is 0. For example. . One interpretation of entropy from information theory is that it specifies the minimum number of bits of information needed to encode the classification of an arbitrary member of S (i. In all calculations involving entropy we define 0log0 to be 0. if all members are positive (pp= 1). and the entropy is 0. called entropy. Note the entropy is 1 (at its maximum!) when the collection contains an equal number of positive and negative examples. Which attribute is the best classifier? The estimation criterion in the decision tree algorithm is the selection of an attribute to test at each decision node in the tree.3 shows the form of the entropy function relative to a binary classification. including 15 positive and 10 negative examples [15+. To illustrate. A good quantitative measure of the worth of an attribute is a statistical property called information gain that measures how well a given attribute separates the training examples according to their target classification. one bit is required to indicate whether the drawn example is positive or negative. as p+ varies between 0 and 1. containing only positive and negative examples of some target concept (a 2 class problem). the receiver knows the drawn example will be positive. binary classification is defined as: Entropy (S) = – pp log2 pp – pn log2 pn where pp is the proportion of positive examples in S and pn is the proportion of negative examples in S.26 | Data Mining and Warehousing 4. A leaf of the tree specifies the expected value of the goal attribute for the records described by the path from the root to that leaf. Entropy—a measure of homogeneity of the set of examples In order to define information gain precisely.1 Basic Ideas Behind ID3 In the decision tree each node corresponds to a non-goal attribute and each arc to a possible value of that attribute.5. On the other hand. Then the entropy of S relative to this classification is Entropy (S) = – (15/25) log2 (15/25) – (10/25) log2 (10/25) = 0.e. we need to define a measure commonly used in information theory. suppose S is a collection of 25 examples. 4. 10–]. so no message need be sent. then pn is 0. If the collection contains unequal numbers of positive and negative examples. if pp is 1. a member of S drawn at random with uniform probability). [This defines what a decision tree is. that characterizes the (im)purity of an arbitrary collection of examples. and Entropy (S) = –1′ log2(1) – 0′ log20 = –1′ 0 – 0′ log20 = 0.. the entropy is between 0 and 1. Entropy is used to measure how informative is a node.] In the decision tree at each node should be associated the non-goal attribute which is most informative among the attributes not yet considered in the path from the root. For example. Given a set S.970 Notice that the entropy is 0 if all members of S belong to the same class. Fig. The goal is to select the attribute that is most useful for classifying examples. the entropy of set S relative to this simple. This measure is used to select among the candidate attributes at each step while growing the tree.4.2.

The expected entropy described by this second term is simply the sum of the entropies of each subset Sv. Sv = {s ∈ S | A(s) = v}). then a collection of messages can be encoded using on average less than 1 bit per message by assigning shorter codes to collections of positive examples and longer codes to less likely negative examples. A ) = Entropy ( S ) − v ∈ Values ( A) ∑ Sv S Entropy ( Sv ) where Values(A) is the set of all possible values for attribute A. More precisely. A) of an attribute A.4 Pp Fig. The measure we will use. is simply the expected reduction in entropy caused by partitioning the examples according to this attribute.8 Entropy 0. Information gain measures the expected reduction in entropy Given entropy as a measure of the impurity in a collection of training examples. Note the logarithm is still base 2 because entropy is a measure of the expected encoding length measured in bits. the information gain. relative to a collection of examples S. If the target attribute takes on c different values.6 0. weighted by the fraction of examples |Sv|/|S| . Gain (S. and Sv is the subset of S for which attribute A has value v (i.3 The entropy function relative to a binary classification. called information gain. Thus far we have discussed entropy in the special case where the target classification is binary.Data Mining Techniques | 27 1 0. is defined as Gain ( S. we can now define a measure of the effectiveness of an attribute in classifying the training data.2 0 0 0. as the proportion of positive examples pp varies between 0 and 1 0.8. then the entropy of S relative to this cwise classification is defined as Entropy ( S ) = ∑ − pi log 2 pi i =1 c where pi is the proportion of S belonging to class i.4 0. the maximum possible entropy is log2c.e. Note also that if the target attribute can take on c possible values. 4.6 0.8 1 If pp is 0.2 0.. Note the first term in the equation for Gain is just the entropy of the original collection S and the second term is the expected value of the entropy after S is partitioned using attribute A.

The value of Gain(S. by knowing the value of attribute A. .. . or 2.e. S: a training set) returns a decision tree. S2).A) is the number of bits saved when encoding the target value of an arbitrary member of S. return a single node with value failure. C... Let { S j | j=1.ID3(R-{A}. This process continues for each new leaf node until either of two conditions is met: 1...28 | Data Mining and Warehousing that belong to Sv. . Attributes that have been incorporated higher in the tree are excluded.. given the value of some other attribute A... am going respectively to the trees (ID3(R-{A}. Algorithm function ID3 Input: (R: a set of non-target attributes. [in that case there may be errors. Gain (S. S1). Let A be the attribute with largest Gain(A. Every attribute has already been included along this path through the tree. the training examples associated with this leaf node all have the same target attribute value (i.2. m} until they are empty end. this time using only the training examples associated with that node. If R is empty.. C. Put another way. m} be the values of attribute A. Gain (S. The process of selecting a new attribute and partitioning the training examples is now repeated for each non-terminal descendant node. a2. begin If S is empty. Return a tree with root labeled A and arcs labeled a1.... m} be the subsets of S consisting respectively of records with value aj for A. then return a single node with the value of the most frequent of the values of the target attribute that are found in records of S.A) is therefore the expected reduction in entropy caused by knowing the value of attribute A. .2.2. so that any given attribute can appear at most once along any path through the tree. ... C: the target attribute. . C. ID3(R-{A}.A) is the information provided about the target attribute value. examples that will be improperly classified]. return a single leaf node with that value. Sm). If S consists of records all with the same value for the target attribute. Recursively apply ID3 to subsets { S j | j=1. Let { A j | j=1.S) among attributes in R. their entropy is zero).

strong} Examples of set S are: TABLE 4. overcast. The range of entropy is 0 (“perfectly classified”) to 1 (“totally random”). The target classification is “should we play cricket” which can be yes or no. Entropy(S) = –(9/14) Log2 (9/14) – (5/14) Log2 (5/14) = 0. Gain(S. A) is information gain of example set S on attribute A is defined as Gain(S. The weather attributes are outlook. mild. They can have the following values: outlook = {sunny. A) = Entropy(S) – Σ ((|Sv|/|S|) * Entropy(Sv)) Where: Σ = value v of all possible values of attribute A S v = subset of S for which attribute A has value v |S v| = number of elements in Sv |S| = number of elements in S.940 Notice entropy is 0 if all members of S belong to the same class (the data is perfectly classified). Over the course of 2 weeks. humidity. .2 Weather Report Day D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 D11 D12 D13 D14 Outlook Sunny Sunny Overcast Rain Rain Rain Overcast Sunny Sunny Rain Sunny Overcast Overcast Rain Temperature Hot Hot Hot Mild Cool Cool Cool Mild Cool Mild Mild Mild Hot Mild Humidity High High High High Normal Normal Normal High Normal Normal Normal High Normal High Wind Weak Strong Weak Weak Weak Strong Strong Weak Weak Weak Strong Strong Weak Strong Play ball No No Yes Yes Yes No Yes No Yes Yes Yes Yes Yes No If S is a collection of 14 examples with 9 YES and 5 NO examples then.Data Mining Techniques | 29 Example of ID3 Suppose we want ID3 to decide whether the weather is amenable to playing cricket.2). data is collected to help ID3 build a decision tree (see Table 4. normal} wind = {weak. and wind speed. temperature. rain} temperature = {hot. cool} humidity = {high.

029 Gain(S. Humidity) = 0.048 Entropy(Sweak) = – (6/8)*log2(6/8) – (2/8)*log2(2/8) = 0. the gain is calculated and the highest gain is used in the decision node. we only decide on the remaining three attributes: Humidity. Ssunny = {D1. The gain is calculated for all four attributes: Gain(S.00 For each attribute. Temperature) = 0. The next question is “what attribute should be tested at the Sunny branch node?” Since we have used Outlook at the root. 6 of the examples are YES and 2 are NO. D2. D9. Since Outlook has three possible values.30 | Data Mining and Warehousing Suppose S is a set of 14 examples in which one of the attributes is wind speed. suppose there are 8 occurrences of Wind = Weak and 6 occurrences of Wind = Strong. The values of Wind can be Weak or Strong. Gain(S. Temperature. We need to find which attribute will be the root node in our decision tree. rain). D11} = 5 examples from Table 1 with outlook = sunny Gain(Ssunny.246 Gain(S.151 Gain(S. For Wind = Strong. Therefore.4 A decision tree for the concept Playball . therefore it is used as the decision attribute in the root node. 4.940 – (8/14)*0.048 Outlook attribute has the highest gain. overcast. Outlook) = 0. For attribute Wind.811 Entropy(Sstrong) = – (3/6)*log2(3/6) – (3/6)*log2(3/6) = 1.Wind) = Entropy(S)–(8/14)*Entropy(Sweak)–(6/14)*Entropy(Sstrong) = 0. For Wind = Weak. or Wind. Wind) = 0. The classification of these 14 examples are 9 YES and 5 NO. D8.00 = 0.811 – (6/14)*1. Humidity) = 0. 3 are YES and 3 are NO. the root node has three branches (sunny.970 Outlook Sunny Humidity Overcast Yes Rain Wind High Normal Strong Weak No Yes No Yes Fig.

All non-terminal nodes contain splits.4 A Detailed Survey on Decision Tree Technique A decision tree contains zero or more internal nodes and one or more leaf nodes. Some specific applications include medical diagnosis. Each leaf node has a class label associated with it. ID3 has been incorporated in a number of commercial rule-induction packages. The final decision = tree The decision tree can also be expressed in rule format: IF outlook = sunny AND humidity = high THEN cricket = no IF outlook = rain AND humidity = high THEN cricket = no IF outlook = rain AND wind = strong THEN cricket = yes IF outlook = overcast THEN cricket = yes IF outlook = rain AND wind = weak THEN cricket = yes. The easiest situation for decision tree learning occurs when each attribute takes on a small number of disjoint possible values (e.Data Mining Techniques | 31 Gain(Ssunny. classification of soybean diseases. equipment malfunctions by their cause. Instances are described by a fixed set of attributes (e. Quinlan Quinlan Agrawal.g. cold).4. 4.6 Decision Tree Algorithms Name of the Algorithm Developer Year CHAID CART ID3 C4.. 4. Agrawal. et al. which test the value of a mathematical or logical expression of the attributes.4. All internal nodes have two or more child nodes. This process goes on until all data is classified perfectly or we run out of attributes. The task of constructing a tree from the training set is called tree induction. it is used as the decision node. et al. et al.019 Humidity has the highest gain.570 Gain(Ssunny. credit risk assessment of loan applications. 4. therefore.5 SLIQ SPRINT Kass Breimarn. hot). 1980 1984 1986 1993 1996 1996 . and web search classification.5 Appropriate Problems for Decision Tree Learning Instances are represented by attribute-value pairs. mild. temperature) and their values (e.4. hot... Wind) = 0. Temperature) = 0.g.g.

Trouble with Non-Rectangular Regions. 4. 1 0 0 0 1 1 1 1 1 1 1 0 0 0 1 0 0 0 0 0 0 1 1 1 1 1 1 1 Fig. For instance. the reproduction process goes through Crossover and Mutation operations. Decision trees are also problematic for time-series data unless a lot of effort is put into presenting the data in such a way that trends and sequential patterns are made visible. Computationally Expensive to Train. or interest rate.1 Crossover Using simple single point crossover reproduction mechanism. one crossover point is selected and the string from the beginning of chromosome to the crossover point is copied from one parent and the rest is copied from the second parent resulting in a child. consider the following chromosomes and crossover point at position 4. The weaknesses of decision tree methods Decision trees are less appropriate for estimation tasks where the goal is to predict the value of a continuous variable such as income.7 Strengths and Weaknesses of Decision Tree Methods strengths of decision tree methods Decision trees are able to generate understandable rules. 4. 2. a point along the chromosome length is randomly chosen at which the crossover is done as illustrated. blood pressure. (a) Single Point Crossover: In this type. . 1. 4. 3.5. Decision trees provide a clear indication of which fields are most important for prediction or classification. 4. Error-Prone with Too Many Classes. Decision trees are able to handle both continuous and categorical variables. 2.5 GENETIC ALGORITHM Genetic Algorithms/Evolutionary algorithms are based on Darwin’s theory of survival of the fittest. Two programs called chromosomes combine to produce a third program called child.5 Single point crossover Here we observe that the regions after the crossover point are interchanged in the children.4.32 | Data Mining and Warehousing 4. Here the best program or logic survives from a pool of solutions. 3. The 1. Decision trees perform classification without requiring much computation.

two crossover points are selected and the string from beginning of one chromosome to the first crossover point is copied from one parent. Parent A Parent B Offspring x + + x = x y 3 y 2 y 2 Fig. mutation involves randomly generating a new character in a specified position.2 Mutation As new individuals are generated. mutation may be done by flipping a bit. its behavior changes radically. the part from the first to the second crossover point is copied from the second parent and the rest is copied from the first parent.7 Tree crossover 4.6 Two point crossover Here we observe that the crossover regions between the crossover points are interchanged in the children. when used in the GA. One crossover point is chosen at random and parents are divided in that point and parts below crossover point are exchanged to produce new offspring. For instance. 4. However. 3 2 1 5 6 7 4 2 7 6 5 1 4 3 3 2 6 5 1 7 4 2 7 1 5 6 4 3 Fig.5. each character is mutated with a given probability. 4. When used by itself without any crossover. This type of crossover is mainly employed in permutation encoding and value encoding where a single point crossover would result in inconsistencies in the child chromosomes. mutation is equivalent to a random search consisting of incremental random modification of the existing solution and acceptance if there is improvement. while in a non-binary-coded GA. consider the following chromosomes and crossover points at positions 2 and 5. In the GA. Mutation produces incremental random changes in the offspring generated through crossover. (c) Tree Crossover: The tree crossover method is most suitable when tree encoding is employed. mutation serves the crucial role for replacing the gene values lost from the population during the selection process so that they can . In a binary-coded Genetic Algorithm.Data Mining Techniques | 33 (b) Two Point Crossover: In this type.

. 4. * x + x + y 3 y 3 Fig. Here. consider mutation at allele 4. For instance.86 4.29 5. 4. 1 0 0 0 0 1 1 0 0 1 0 1 Fig. 1 2 3 4 5 6 1 6 3 4 5 2 Fig.55) => (1. a bit is changed from 0 to 1 and vice-versa. For instance. (1.34 | Data Mining and Warehousing be tried in a new context. deals with finding a structure in a collection of unlabeled data.e.55) (d) Operator Manipulation: This method involves changing the operators randomly in an operator tree and hence is used with tree-encoded problems. Hence. The type of mutation used as in crossover is dependent on the type of encoding employed. or of providing the gene values that were not present in the initial population. . two random points in the chromosome are chosen and interchanged. Here.6 CLUSTERING 4. this method is specifically useful for real value encoded problems.8 Bit inversion (b) Order Changing: This type of mutation is specifically used in permutation-encoded problems.9 Order changing (c) Value Manipulation: Value manipulation refers to selecting random point or points in the chromosome and adding or subtracting a small number from it. The various types are as follows: (a) Bit Inversion: This mutation type is employed for a binary encoded problem. 4.68 2.1 What is Clustering? Clustering can be considered the most important unsupervised learning problem.73 4. a bit is randomly selected and inverted i.11 5. so. 4.6.68 2.10 Operator manipulation Here we observe that the divide operator in the parent is randomly changed to the multiplication operator. as every other problem of this kind. For instance.22 5.29 5.

Data Mining Techniques | 35 Definition of clustering could be “the process of organizing objects into groups whose members are similar in some way”. not according to simple similarity measures. j ) = xi 1 − x j 1 + xi 2 − x j 2 + . + xip − x jp 2 2 2 Manhattan Distance Function d ( i. A cluster is therefore a collection of objects which are “similar” between them and are “dissimilar” to the objects belonging to other clusters. xi2. xip ) and j = (xj1. the following common distance functions can be defined: Euclidean Distance Function d ( i.... 4. we easily identify the 4 clusters into which the data can be divided. Another kind of clustering is conceptual clustering: two or more objects belong to the same cluster if this one defines a concept common to all that objects. + xip − x jp .. In other words. objects are grouped according to their fit to descriptive concepts.6. 4. xjp ). xj2... the similarity criterion is distance: two or more objects belong to the same cluster if they are “close” according to a given distance (in this case geometrical distance). This is called distance-based clustering. .. .. Graphical Example Fig.2 Distance Functions Given two p-dimensional data objects i = (xi1.11 Clustering process-example In the example. j ) = xi 1 − x j 1 + xi 2 − x j 2 + ...

in finding “natural clusters” and describe their unknown properties (“natural” data types). Nonhierarchical and 2. To prevent this problem. for instance: ² Marketing: finding groups of customers with similar behavior given a large database of customer data containing their properties and past buying records. ² Libraries: book ordering. ² Biology: classification of plants and animals given their features.3 Goals of Clustering The goal of clustering is to determine the intrinsic grouping in a set of unlabeled data. ² WWW: document classification.6. For instance. such as the k-means algorithm.36 | Data Mining and Warehousing When using the Euclidean distance function to compare distances. the relationship between clusters is undetermined. ² Earthquake studies: clustering observed earthquake epicenters to identify dangerous zones. ² Insurance: identifying groups of motor insurance policy holders with a high average claim cost.5 Types of Clustering Algorithms There are two types of clustering algorithms: 1. ² City-planning: identifying groups of houses according to their house type. in finding useful and suitable groupings (“useful” data classes) or in finding unusual data objects (outlier detection). so when using the Euclidean distance function. identifying frauds.4 Applications Clustering algorithms can be applied in many fields. 4. so that clusters can be formed from objects with a high similarity to each other.6.6. in such a way that the result of the clustering will suit their needs. With both of these approaches. we could be interested in finding representatives for homogeneous groups (data reduction). Hierarchical clustering repeatedly links pairs of clusters until every data object is included in the hierarchy. But how to decide what constitutes a good clustering? It can be shown that there is no absolute “best” criterion which would be independent of the final aim of the clustering. for two distances. an important issue is how to determine the similarity between two objects. Consequently. In nonhierarchical clustering. 4. 4. If some of an object’s attributes are measured along different scales. d1 and d2. clustering weblog data to discover groups of similar access patterns. . √d1 > √d2 ⇔ d1 > d2. value and geographical location. the attribute values are often normalized to lie between 0 and 1. Hierarchical. it is not necessary to calculate the square root because distances are always positive numbers and as such. attributes with larger scales of measurement may overwhelm attributes measured on a smaller scale. it is the user which must supply this criterion.

6. These means are used as the new cluster points and each object is reassigned to the cluster that it is most similar to. The value of k may or may not be specified and a clustering criterion. Select a clustering criterion. The k-means algorithm initializes k clusters by arbitrarily selecting one object to represent each cluster. typically the squared-error criterion. or clusters.5  0  1   1   0  0.13 Sample data—k-means clustering process .12 k-means algorithm Example 0 0  1  1 0. Initialize cluster centers with those k clusters. 3. k-means Algorithm: 1.1 Nonhierarchical Algorithms K-means Algorithm The k-means algorithm is one of a group of algorithms called partitioning methods. determine a partition of the objects into k groups.Data Mining Techniques | 37 4. must be adopted. The solution to this problem is straightforward. (b) Compute new cluster centers as mean value of the objects in each cluster until no change in cluster center calculation. 4.12. 2. This continues until there is no longer a change when the clusters are recalculated. The algorithm is shown in Fig. Recall that a partition divides a set into disjoint parts that together include all members of the set. Fig. Do loop (a) Partition by assigning or reassigning all data objects to their closest cluster center. 4. such that the objects in a cluster are more similar to each other than to objects in different clusters.5  5  6   6   5  5. The problem of partitioned clustering can be formally stated as follows: Given n objects in a d-dimensional metric space. 4. Select k clusters arbitrarily.5 Pi =  5 5  6  6 5. then for each data object select the cluster that optimizes the criterion. Each of the remaining objects is assigned to a cluster and the clustering criterion is used to calculate the cluster mean.5   * * * * * * * * * * Fig.5.

2.0. so algorithm is finished.1)}.1.5)/7 = 4.0).5). The seed points could be the first k objects or k objects chosen randomly from the input objects. The detailed computation is as follows: Step 1: Initialize k partitions.(1. for the third object: dis(1.6).5.5. After calculating the distance for all points.(0.(1.5.5) differ from old centers C1 = (0.1). 4.16) and C2 = (4.1). that point is the cluster center.5)/7 = 4.5) New center for C2 = (5.2) differ from old centers C1={(0. Step 2: Since there is only one point in each cluster.5)} and C2 = {(0. Step 3b: Compute new cluster center for each cluster.(5. Algorithm is done: Centers are the same as in Step 3b´.5.6).2) (0+1+5+5+6+6+5. so this object is assigned to C2.(5.1).16) (0+1+0. assigning the object to the closest cluster.(6.5. Step 3a″: New centers C1 = (0.5). so the loop is repeated.(5.4.3) = |1–0| + |1–1| = 1.(5.5. The fifth object is equidistant from both clusters.2). Result is same as in 3b´. Reassign the ten objects to the closest cluster center.6). so the loop is repeated.5).0. Step 3b″: Compute new cluster centers.5.5.1)}.38 | Data Mining and Warehousing Suppose we are given the data in Fig. resulting in: C1 = {(0.0.5)/3 = 0.5.(1. We choose the first two objects as seed points and initialize the clusters as C1 = {(0. the clusters contain the following objects: C1 = {(0.(6.0)} and C2={(0.0.(6.5.0).5.6).0.5.(0.0.16 New center for C2 = (4.5) and C2 = (5.3) = |1–0| + |1–0| = 2 and dis(2.1.1 (1+1+5+5+6+6+5.4.0). .0. An initial partition can be formed by first specifying a set of k seed points.0)} and C2 = {(0.(6.0). Step 3b′: Compute new cluster center for each cluster New center for C1 = (0.13 as input and we choose k = 2 and the Manhattan distance function dis = |x2–x1|+|y2–y1|.(1.5).(0.5)/3 = 0.1).5. so we arbitrarily assign it to C1.5. Step 3a: Calculate the distance between each object and each cluster center.1.5.5)}.5)} C2 = {(5.16) and C2 = (4. For example. New center for C1 = (0.5 (0+0+0.5).4.(5.5)}. Step 3a′: New centers C1 = (0.

The distance function in this algorithm can determine similarity of clusters through many methods. These groups are successively combined based on similarity until there is only one group remaining or a specified termination condition is satisfied. including single link and group-average.. cluster) and calculates the distance between clusters as the distance between the average values.6. Single link calculates the distance between two clusters as the shortest distance between any two objects contained in those clusters. 4.15 shows a simple hierarchical algorithm. it can also cause problems if an erroneous merge is done. the hierarchy graph is called a dendogram. For n objects. merge points need to be chosen carefully. 4. Thus. that is top-down or bottomup.2 Hierarchical Clustering Hierarchical clustering creates hierarchy of clusters on the data set. 4. Hierarchical algorithms can be either agglomerative or divisive.Data Mining Techniques | 39 4. Each object in X is initially used to create a cluster containing a single object. it cannot be undone. or k = 1. After the hierarchy is built. Hierarchical algorithms are rigid in that once a merge has been done. the number of clusters in C decreases until there is only one cluster remaining.5. The hierarchy of clusters is implicitly represented in the nested sets of C. This hierarchical tree shows levels of clustering with each level having a larger number of smaller clusters. . from 1 to n. the user can specify the number of clusters required. All agglomerative hierarchical clustering algorithms begin with each object as a separate group. When a pair of clusters is merged. the number of clusters (k) is not specified in hierarchical clustering.14 Sample dendogram without clustering process In the context of hierarchical clustering. The top level of the hierarchy represents one cluster.e. Although there are smaller computational costs with this. Fig. These clusters are successively merged into new clusters. To examine more clusters. Here we describe a simple agglomerative clustering algorithm. As such. f l c e h a k b i d m g j Fig. C. Group-average first finds the average values for all objects in the group (i. n-1 mergings are done. the original clusters are removed from C. Fig. we simply need to traverse down the hierarchy. which are added to the set of clusters. Unlike with the k-means algorithm. containing all the objects from X.14 shows a sample dendogram that could be produced from a hierarchical clustering algorithm.

. 4.{x7}.cb } 3... where x1 = (0.cmin2) = minimum dis(ci .0).{x9}. where ci is a member of the set of clusters C. each element xi of X is placed in a cluster ci . shown in Fig.xn} A distance function dis(c1. for i = 1 to n ci = {xi } end for 2. The set X contains n = 10 elements.40 | Data Mining and Warehousing Given: A set X of objects {x1..{x5}. 4. We will use the Manhattan distance function and the single link method for calculating distance between clusters.cj) for all ci . {x4}. C = {c1.{x2 }. 4.16 represented in matrix and graph form. while C.{x6}. 0 1  3  2 6 X= 6 5  3  0 2  0 1  1  4 3  6 2  5  2 1  * * * * * * * * * * Fig..cmin2} to C (d) l = l + 1 end while Fig.c2) 1.16 Sample data Step 1: Initially. {x 8}.. .. C = {{x1}.cj in C (b) remove cmin1 and cmin2 from C (c) add {cmin1.15 Agglomerative hierarchical algorithm Example: Suppose the input to the simple agglomerative algorithm described above is the set X.size > 1 do (a) (cmin1. l = n+1 4.{x10 }} Step 2: Set l = 11.. x1 to x10 .{x3 }.

c15) ² C = {{{x4}. between c3 and c11 because the distance between x3 and x10 is 1.size = 8 ² (cmin1.{x 3}. This occurs between.{x8}}. {x10}}.{x7}}}} Step 3: (Eighth iteration) C. ² Add c12 to C.{x9}. Step 3: (Third iteration) C.c12) ² C = {{x4}.{x7}}} Step 3: (Sixth iteration) C.{{x2}.c min2) = (c5.{x 5}.cmin2) = (c3.{x1}}} Step 3: (Fourth iteration) C. {x9}.c8) ² C = {{x 5}.{x7}.{x6}.{{x4}. {x 10}}. {{x 5}.{x6}. {{x5}.size = 10 ² The minimum single link distance between two clusters is 1. (cmin1.{x6}.cmin2) = (c6. Step 3: (Second iteration) C. {x8}.c10) ² Since l = 10.{{{{x2}. between c2 and c10 and between c3 and c10.cmin2) = (c9.{x1}}.cmin2) = (c1. where x10 is in c11. {x4}. {x 10}}} ² Set l = l + 1 = 12. {x 8}.{x10}}. {x3}}.c11) ² c12 = c3 U c11 = {{{x2}.{{x4}. This occurs in two places. {x3}}.{x 1}}. {x9}.size = 5 ² (cmin1. (cmin1. {x10}}.size = 4 ² (cmin1.{{{{x2}.c7) ² C = {{x6}.{x5}.c min2) = (c4.{x1}}. C = {{x1}.cmin2) = (c 2. {x 3}}. {x10}}. {{{{{x 2}.{x 7}}. {x10}}.size = 7 ² (cmin1.{{x6}. {x3}}. {x3}}} ² Set l = 13.size = 6 ² (cmin1. {x8}. ² Add c11 to C. c11 = c2 U c10 = {{x2}. {x 3}}.c16) . {{x4}. C = {{x1}.size = 9 ² The minimum single link distance between two clusters is 1.{x6}. {x10}}. {x4}.{x8}}.{x9}}.{x1}}.{{{{x2}. Depending on how our minimum function works we can choose either pair of clusters.{x7}.Data Mining Techniques | 41 Step 3: (First iteration of while loop) C.{x9}.{x9}.c min2) = (c14.{x7}.{x 8}}. Arbitrarily we choose the first.size = 3 ² (cmin1.{{{x2}.{x5}.{x9}}} Step 3: (Seventh iteration) C. {{x5}.{x3}} ² Remove c3 and c11 from C.{{{{{x2}.c13) ² C = {{x6}.{x10}} ² Remove c2 and c10 from C.{x7}.{x 8}}} Step 3: (Fifth iteration) C.

{{{{{x2}.42 | Data Mining and Warehousing ² C = { {{x6}.16 are grouped together more closely in the hierarchy. 4. {{{{{x2}.c18) ² C = {{{{x4}. {{x5}.18 Sample dendogram after clustering . The corresponding dendogram formed from the hierarchy in C is shown in Fig.{x8}}.{x1}}.c min2) = (c17. {{x5}.{x7}}}. {x10}}. 4. * * * * * * * * * * Fig.17.17 Graph with clusters X6 X5 X7 X1 X2 X10 X3 X9 X4 X8 Fig.18. {{x6}. 4. {x 10}}.{x8}}. {x3}}. Algorithm done.{x7}}}} Step 3: (Tenth iteration) C. {{{x4}.{x9}}}.{x 9}}}} Step 3: (Ninth iteration) C. 4.size = 1. The points which appeared most closely together on the graph of input data in Fig.size = 2 ² (cmin1. {x3}}. The cluster created from this algorithm can be seen in Fig.{x1}}. 4.

2. ² Find the document not in the MST that is closest to any point in the MST. and add it to the current MST fragment. ² If a document remains that is not in the MST fragment.6. SLINK Algorithms ² ² ² ² ² Array record cluster pointer and distance information Process input documents one by one.Data Mining Techniques | 43 4. ² Time: Voorhees alg. so have small. Distance between 2 clusters is the distance between the closest pair of points. ² Space: O(N). worst case is O(N**3) but sparse matrices require much less ² Space: Voorhees alg. widely used. Complete Link Method ² Similarity: Join least similar pair between each of two clusters ² Type of clusters: All entries in a cluster are linked to one another within some minimum similarity. using the matrix Relabel clusters. repeat the previous step. chains. ² Disadvantages: Difficult to apply to large data sets since most efficient algorithm is general Hierarchical agglomerative clustering method (HACM) using stored data or stored matrix approach. worst case is O(N**2) but sparse matrices require much less ² Advantages: Good results from (Voorhees) comparative studies. tightly bound clusters. to create an initial MST fragment. . each of which is in one of the two clusters. For each: Compute and store a row of the distance matrix Find nearest other point. so no need arises to recalculate the similarity matrix. ² Type of clusters: Long straggly clusters. ² Time: Usually O(N**2) though can range from O(N log N) to O(N**5). No cluster centroid or representative required. Single Link Method ² Similarity: Join the most similar pair of objects that are not yet in the same cluster. ² Advantages: Theoretical properties. ellipsoidal. ² Prim-Dijkstra algorithm operates as follows: ² Place an arbitrary document in the MST and connect its NN (Nearest Neighbor) to it.3 Other Hierarchical Clustering Methods 1.5. as needed MST Algorithms ² MST has all info needed to generate single link hierarchy in O(N**2) operations. efficient implementations. or the single link hierarchy can be built at the same time as MST. ² Disadvantages: Unsuitable for isolating spherical or poorly separated clusters.

Save space and time if use inner product of 2 (appropriately weighted) vectors — see Voorhees alg. Algorithm . in O(N) space Ward’s Method (minimum variance method) ² Similarity: Join the cluster pair whose merger minimizes the increase in the total withingroup error sum of squares. based upon all objects in the cluster. poor at recovering elongated clusters RNN: We can apply a reciprocal nearest neighbor (RNN) algorithm. . based on Euclidean distance between centroids ² Type of clusters: Homogeneous clusters in a symmetric hierarchy.Single Cluster: 1. two clusters of size mi and mj can be merged as soon as the mi × mj th similarity of documents in those clusters is reached. yields unique and exact hierarchy ² Disdvantages: Sensitive to outliers. Follow the NN chain from it till an RNN pair is found. ² Requires sorted list of doc-doc similarities ² Requires way to count no. below. can use centroids only to compute cluster-cluster sim. cluster centroid nicely characterized by its center of gravity ² Time: O(N**2) ² Space: O(N) — carries out agglomerations in restricted spatial regions ² Advantages: Good at recovering cluster structure. ² Type of clusters: Intermediate in tightness between single link and complete link ² Time: O(N**2) ² Space: O(N) ² Advantages: Ranked well in evaluation studies ² Disadvantages: Expensive for large collections Algorithm ² Sim between cluster centroid and any doc = mean sim between the doc and all docs in the cluster. since for any point or cluster there exists a chain of nearest neighbors (NNs) that ends with a pair that are each others’ NN. ² Hence.44 | Data Mining and Warehousing Voorhees Algorithm ² Variation on the sorted matrix HACM approach ² Based on fact: if sims between all pairs of docs are processed in descending order. of sims seen between any 2 active clusters 3. Select an arbitrary doc 2. Group Average Link Method ² Similarity: Use the average value of the pairwise links within a cluster.

return to step 1. Identify the 2 closest points and combine them into a cluster 2. and each HACM can be characterized by its own set of Lance-Williams parameters. but one need not store the data set. ² Disadvantages: Newly formed clusters may differ overly much from constituent points. using the appropriate Lance-Williams dissimilarities. O(N**2 log N**2) to sort it. to calculate dissimilarities between a new cluster and existing points. 4. but will be O(N**3) if matrix is scanned linearly. Identify and combine the next 2 closest points (treating existing clusters as points too) 3. Merge the 2 points in that pair. Storage is therefore O(N**2) and time is at least O(N**2). and the matrix can be processed linearly. ² Type of clusters: Cluster is represented by the coordinates of the group centroid or median. General algorithm for HACM All hierarchical agglomerative clustering methods (HACMs) can be described by the following general algorithm.FASMI) refers to technology that allows users of multidimensional databases to . Algorithm 1. ² Stored data approach: O(N) space for data but recompute pairwise dissimilarities so need O(N**3) time ² Sorted matrix approach: O(N**2) to calculate dissimilarity matrix.Data Mining Techniques | 45 3. and then apply Lance-Williams to recalculate dissimilarities between cluster centers. replacing them with a single new point. Centroid and Median methods ² Similarity: At each stage join the pair of clusters with the most similar centroids. 4. O(N**2) to construct hierarchy. This formula has 3 parameters. which reduces disk accesses. if there is a point in the NN chain preceding the merged points. return to step 2 else to step 1. 4. stop. If there is only one point. meaning that updates may cause large changes throughout the cluster hierarchy. based on the dissimilarities prior to forming the new cluster. How to apply general algorithm? Lance and Williams proposed the Lance-Williams dissimilarity update formula. Implementations of the general algorithm ² Stored matrix approach: Use matrix.OLAP (or Fast Analysis of Shared Multidimensional Information . Then. If more than one cluster remains. the algorithm above can be applied. Otherwise.7 ONLINE ANALYTIC PROCESSING (OLAP) The term On-Line Analytic Processing .

g. OLAP facilities can be integrated into corporate (enterprise-wide) database systems and they allow analysts and managers to monitor the performance of the business (e. they may involve seasonal adjustments.e.46 | Data Mining and Warehousing generate on-line descriptive or comparative summaries (“views”) of data and other analytic queries. Discovery of association rules are showing attribute-value conditions that occur frequently together in a given set of data..g. The final result of OLAP techniques can be very simple (e. This above condition extracts the hidden information i. In the . let T be the set of all “baskets” or “carts” of products bought by the customers from a supermarket – say on a given day. There are two types of Association rule levels. descriptive statistics. he will also buy butter as side dish.. contain dynamically updated information) through efficient “multidimensional” queries that reference various types of data. The support of an item set S is the percentage of those transactions in T which contain S. The minimum percentage of instances in the database that contain all items listed in a given association rule. such as various aspects of the manufacturing process or numbers and types of completed transactions at different locations) or the market.1 Rules for Support Level The minimum percentage of instances in the database that contain all items listed in a given association rule. if a customer used to buy bread.g. More information about OLAP is discussed in Chapter 6. removal of outliers. obviously. Support of an item set Let T be the set of all transactions under consideration. 4.. the term applies to analyses of multidimensional databases (that may. simple cross-tabulations) or more complex (e. e. and other forms of cleaning the data). The set of items a customer buys is referred to as an item set.g. frequency tables. Typically the relationship will be in the form of a rule: IF {bread} THEN {butter}. Market Basket Analysis is a modeling technique based on the theory that if you buy a certain group of items then you are more (or less) likely to buy another group of items.8.8 ASSOCIATION RULES Association rule mining finds interesting associations and/or correlation relationships among large set of data items. and market basket analysis seeks to find relationships between purchases. Support Level Confidence Level 4. analyses referred to as OLAP do not need to be performed truly “online” (or in real-time). Association rules show attributes value conditions that occur frequently together in a given dataset. A typical and widely-used example of association rule mining is Market Basket Analysis... Note that despite its name.

the confidence of a rule is the number of cases in which the rule is correct relative to the number of cases in which it is applicable. A transaction is a set of items (t. Shortly after that. then support (S) = (242/318) = 76.3 APRIORI Algorithm The first algorithm to generate all frequent sets and confident association rules was the AIS algorithm by Agrawal et al. ∈. rule confidence is the conditional probability that B is true when A is known to be true. If there are 318 customers and 242 of them buy such a set U or a similar one that contains S. An association rule is like an implication: X –> Y means that if item set X occurs in a transaction. An item set is frequent if its support is greater or equal to a threshold value. bread. milk}. Confidence (R) = (support ({A. that contain X). Transactions are often called baskets. B})) *100% More Confidence of an association rule To evaluate association rules. The relative support is the absolute support divided by the number of transactions (i.1%. the algorithm was improved and renamed Apriori by Agrawal et al. C})/support ({A.e. apples. then item set Y also occurs . n). potatoes} then S is obviously a subset of X. Association rule mining problem This program is also capable of mining association rules.2 Rules for Confidence Level “If A then B”.. The (absolute) support or the occurrence of X (denoted by Supp(X) is the number of transactions that are supersets of X (i. sausages.e. if a customer buys the set X = {milk. then Support(S) = (|U|/|T|) *100% where |U| and |T| are the number of elements in U and T. respectively.. 4. which was given together with the introduction of this mining problem. If U is the set of all transactions that contain all items in S. For example. If a customer buys butter and bread. than the rule is not applicable and thus (obviously) does not say anything about this customer.e.. for example S = {bread. For example. cheese. let R = “butter and bread –> milk”.Data Mining Techniques | 47 supermarket example this is the number of “baskets” that contain a given set S of products. referring to the primary application domain (i. If he/she does not buy sugar or does not buy bread or buys neither.8. the confidence of a rule R = “A and B –> C” is the support of the set of all items that appear in the rule divided by the support of the antecedent of the rule. market-basket analysis).e.….. I). by exploiting the monotonic property of the frequency of item sets and the confidence of association rules Frequent item set mining problem A transactional database consists of sequence of transaction: T = (t1. i. 4.. B..8. butter. A set of items is often called the item set by the data mining community. and hence S is in U. banana. onions. then the rule is applicable and it says that he/she can be expected to buy milk.

In the frequent itemset mining problem a transaction database and a relative support threshold (traditionally denoted by min_supp) is given and we have to find all frequent itemsets. Count the support of item sets pruning process ensures candidate sizes are already known to be frequent item sets.48 | Data Mining and Warehousing with high probability. It is like an approximation of p(Y|X). Use the frequent item sets to generate the desired rules. thus conf(X–>Y) = Supp(XUY)/Supp(X). Apriori steps are as follows: Counts item occurrences to determine the frequent item sets. Algorithm Input I D S Output L // Large Itemsets Apriori Algorithm k = 0 // k is used as the scan number L=0 Ci = I // Initial candidates are set to be the items Ck Repeat k = k + 1 Lk = 0 For each Ii ∈ Ck do ci = 0 // initial counts for each itset are 0 for each tj ∈ D do for each Ii ∈ Ck do if Ii ∈ tj then ci = ci +1 for each Ii ∈ Ck do if ci >= (s x |D|) do Lk = Lk U Ii L = L U Lk // Itemsets // Database of transactions // Support . An association rule is valid if its confidence and support are greater than or equal to corresponding threshold values. it is the number of transactions that contain both X and Y divided by the number of transaction that contain X. This probability is given by the confidence of the rule. Candidates are generated.

1 2 1 2 3 2 sup.19 Apriori example . 2 3 3 3 itemset {1 2} {1 3} {1 5} {2 3} {2 5} {3 5} itemset {2 3 5} itemset {2 3 5} Fig. 4. 2 itemset {1} {2} {3} {5} sup. 2 3 3 1 3 itemset {1 2} {1 3} {1 5} {2 3} {2 5} {3 5} sup.Data Mining Techniques | 49 Ck+1 = Apriori_Gen(Lk) Until Ck+1 =0 Input Li – 1 Output Ci // Candidate of size i Apriori-gen algorithm Ci = 0 For each I ∈ Li-1 do For each J ≠ I ∈ Li-1 do If i-2 of the elements in I and J are equal then Ck = Ck U (I U J) // Large Itemsets of size i–1 AR Gen algorithm Input D // Database of transactions I // Items L // Large Itemsets S // Support œ // Confidence Output R // Association rules satisfying s and œ AR Gen Algorithm R=0 For each I ∈ L do For each x C I such that x ≠0 do If support(I)/support(x) >= œ then R=RU(x = > (1 – X)) TID 100 200 300 400 Items 13 4 235 1235 25 itemset {1 3} {2 3} {2 5} {3 5} sup. 2 2 3 2 itemset {1} {2} {3} {4} {5} sup.

Then any one algorithm is used to find the large item sets for the sample. // Missing Large itemsets If ML ≠ 0 then C = Li // set candidates to be the large itemsets Repeat C = C U BD~© // Expand candidate sets using negative border Until no new itemsets are added to C. Confidence 100%) 4. . (2) Original sampling algorithm reduces the number of database scans to one or two. it is viewed to be potentially large in the entire database.8.Ds.50 | Data Mining and Warehousing Rules from example Support(2 3) = 2/4*100 = 50 Support(2 3 5) = 2/4*100 = 50 Confidence(2 3→5) = Support(2 3 5)/support(2 3) = 50/50*100=100% 2 3 → 5 (Support 50%. In an item set is large in a sample.smalls) C = PL U BD~(PL) L=0 For each Ii ∈ C do ci = 0 // Initial counts for each itemset are 0 For each tj ∈ D do // First scan count For each Ii ∈ C do If Ii ∈ tj then ci = ci + 1 For each Ii ∈ C do If ci >= (sx|D|) do L = L U Ii ML = (x| x BD~(PL) x L). Advantages (1) Facilitates efficient counting of item sets in large databases. Input: I // Item set D // Database of transactions S // Support Output: L // Large item set Sampling algorithm: Ds = Sample drawn from Di PL = Apriori(I.4 Sampling Algorithm Here a sample is drawn from the database such that it is memory resident.

Advantages (1) Algorithm is more efficient as it has the advantages of large itemset property wherein a large itemsets must be large in at least one of the partitions.s) C = CULi L=0 For each Ii ∈ C do ci =0 // Initial counts for each item set are 0. (3) It is easier to perform incremental generation of association rules by heating the current style of database as one partition and treating the new entries as a second partitions.….Data Mining Techniques | 51 For each Ii ∈ C do ci = 0 // Initial counts for each itemset are 0 For each tj ∈ D do // Second scan count For each Ii ∈ C do If Ii ∈ tj then ci = ci + 1 If ci … (sx|D| then L = L U Ii 4. Input I // Item set 1 2 D = (D .D .…. D .5 Partitioning Algorithm Here the database D is divided into ‘P’ partitions D1. D2. For each tj ∈ D do // count candidates during second scan For each Ii ∈ C do Ii ∈ tj then ci = ci +1 For each Ii ∈ C do If ci … (sx|D|) then L = L U Ii .Dp) // Database of transactions divided into partition S // Support Output L // Large Itemset Partitioning Algorithm C=0 For I = 1 to p do // Find large itemsets in each partition i Li = Apriori(I. Dn.8. (2) These algorithms adapt better to binted main memory.

1 Web Mining Web mining is mining of data related to the World Wide Web.9. This includes applications to whether. Geographic information systems are used to store information related to Geographic locations on the surface of the Earth. is data mining as applied to spatial databases or spatial data. Data mining activities include prediction of environmental catastrophes. called a snapshot database.9. Each tuple contains the information that is current from the date stored with that tuple to the date stored with the next tuple in temporal order. Spatial Mining or Spatial data mining or knowledge discovery in spatial databases. geology. Applications for spatial data mining are GIS system.9 EMERGING TRENDS IN DATA MINING 4. 4. resource management.52 | Data Mining and Warehousing 4. Content of actual web pages Intra page structure includes the HTML or XML code for the page Inter page structure is the actual linkage structure between web Usage data that describe how Web pages are accessed by visitors User profiles include demographic and registration information obtained about users. Biomedical application including medical imaging and illness diagnosis also require spatial system. environment science. Spatial data are required for many current information technology systems. community infrastructure needs. agriculture.2 Spatial Mining Spatial data are data that a spatial or location component. disaster management. Data are maintained for multiple time points. 4.9.3 Temporal Mining The data that are stored reflect data at a single point in time. not just one time point is called temporal database. Web mining taxonomy The following are the classification of web mining: Web content mining Web page content mining Search result mining Web structure mining Web usage mining General access pattern tracking Customized usage tracking. Web data can be classified into the following classes. Spatial data can be viewed as data about objects that themselves are located in physical space. . and hazardous waste. This may be the data actually present in Web pages or data related to Web activity. medicine and robotics.

3. What is clustering? Explain. 12. 2. 7. Write an essay on emerging trends in data mining.10 DATA MINING RESEARCH PROJECTS 1. What is market-basket analysis? Explain with an example. What is OLAP? Explain the various tools available for OLAP. What is genetic algorithm? Explain. 6. 8. Write short notes on (i) Apriori algorithm. 5. 4. 3. 15. . Describe association rule mining and decision trees. Data Mining in the Insurance Industry Increasing Wagering Turnover Through Data Mining The Transition from Data Warehousing to Data Mining Rule Extraction from Trained Neural Networks Neural Networks for Market Basket Analysis Data Mining for Customer Relationship Management Web Usage Mining using Neural Networks Data Visualization Techniques and their Role in Data Mining Evolutionary Rule Generation for Credit Scoring Improved Decision Trees for Data Mining Data Mining in the Pharmaceutical Industry Data Mining the Malaria Genome Neural Network Modeling of Water-Related Gastroenteritis Outbreaks Data Mining with Self Generating Neural Networks Development of Intelligent OLAP Tools Mining Stock Market Data Data Mining by Structure Adapting Neural Networks Rule Discovery with Neural Networks in Different Concept Levels Cash Flow Forecasting using Neural Networks Neural Networks for Market Response Modeling Advertising Effectiveness Modeling Temporal Effects in Direct Marketing Response EXERCISE 1. (ii) ID3 algorithm. 17. 8. 4. 5. 16. 19. 13. 6. Write in detail about the use of artificial neural network and genetic algorithm for data mining. 9.Data Mining Techniques | 53 4. 11. 20. 7. 18. 10. 21. 22. 14. 2.

data mining tools and techniques have found a niche. They have become the “Hobson’s choice” in a number of industrial applications including.1 APPLICATIONS OF DATA MINING As on this day. there is not a single field where data mining is not applied. ² Ad revenue forecasting ² Churn (turnover) management ² Claims processing ² Credit risk analysis ² Cross-marketing ² Customer profiling ² Customer retention ² Electronic commerce ² Exception reports ² Food-service menu analysis ² Fraud detection ² Government policy setting ² Hiring profiles ² Market basket analysis ² Medical management ² Member enrollment ² New product development ² Pharmaceutical research ² Process control ² Quality control ² Shelf management/store management ² Student recruitment and retention ² Targeted marketing ² Warranty analysis and many more. .CHAPTER 5 REAL TIME APPLICATIONS AND FUTURE SCOPE 5. Starting from marketing down to the medical field.

1.Real Time Applications and Future Scope | 55 The 2003-IBM grading of industries which use data mining are as follows in Table 5. Some of the applications of data mining in this industry are. previously unseen patterns of customer behavior are now coming into clear focus with the aid of new data mining tools. Although banks have employed statistical analysis tools with some success for several years.1 Data Mining in the Banking Sector Worldwide. These statistical tools and even the OLAP find out the answers. but more advanced data mining tools provide insight to the answer. (i) Predict customer reaction to the change of interest rates (ii) Identify customers who will be most receptive to new product offers (iii) Identify “loyal” customers (iv) Pin point which clients are at the highest risk for defaulting on a loan (v) Find out persons or groups who will opt for each type of loan in the following year (vi) Detect fraudulent activities in credit card transactions .1 Industrial Applications of Data Mining Banking Bioinformatics/Biotechnology Direct Marketing/Fundraising eCommerce/Web Entertainment/News Fraud Detection Insurance Investment/Stocks Manufacturing Medical/Pharmacy Retail Scientific Data Security Supply Chain Analysis Telecommunications Travel Others None 1% 2% 4% 2% 1% 8% 13% 10% 10% 5% 1% 9% 8% 3% 2% 6% 6% 9% Industries/fields where data mining is currently applied are as follows: 5.1. banking sector is ahead of many other industries in using mining techniques for their vast customer database. TABLE 5.

5. The many ways in which mining techniques help in direct marketing are. etc. etc. volunteers. (i) Differential market basket analysis helps to decide the location and promotion of goods inside a store and to compare results between different stores..) by. (ii) A predictive market basket analysis can be used to identify sets of item purchases (or events) that generally occur in sequence — something of interest to direct marketers. 5.1.1. and access them from one centralized database. Communication tools such as the Mass E-Mail Manager make it easy to stay in touch with the constituents.g. between customers in different demographic groups. etc. Various applications of data mining techniques in this field are: (i) prediction of structures of various proteins (ii) determining the intricate structures of several drugs (iii) mapping of DNA structure with base to base accuracy 5. voluntary organizations.2 Data Mining in Bio-Informatics and Biotechnology Bio-informatics is a rapidly developing research area that has its roots both in biology and information technology. Techniques followed in data harvesting also help in fund raising (e. (i) Capturing unlimited data on various donors. different seasons of the year. (ii) Setting up multiple funds and default designations based on various criteria such as campaign. between different days of the week.4 Data Mining in Fraud Detection Data dredging has found wide and useful application in various fraud detection processes like (1) Credit card fraud detection using a combined parallel approach . agencies and members of different charitable trusts. criminologists and many others. election fund raising.56 | Data Mining and Warehousing (vii) Predict clients who are likely to change their credit card affiliation in the next quarter (viii) Determine customer preference of the different modes of transaction namely through teller or through credit cards. (iv) Providing a flexible communication and campaign management structure from a single appeal through multiple campaigns. these data are managed and used to build and improve long-term relationships with them. assigning requirements and allowing the system to refer the data to automatically generate awards as a recognition of exceptional donor efforts. to increase company sales.3 Data Mining in Direct Marketing and Fund Raising Marketing is the first field to use data mining tools for its profit. region or membership type to honor the donors’ wishes. prospects.1. relationship management and data mining tools. (iii) Defining award classes. Using powerful search options.

which help companies to reduce costs. As a result it has optimized crop yields. (3) Fraud detection in passport applications by designing a specific online learning diagnostic system. 5. Gone are those days when scientists have to search through reams of printouts and rooms full of tapes to find the gems that make up scientific discovery. This is an example of Temporal mining as it uses a temporal database maintained for multiple time points.1. every activity in telecommunication has used data mining technique. To minimize the resources. retain current customers.1. 5.6 Data Mining in the Insurance Sector Insurance companies can benefit from modern data mining methodologies. increase profits. (4) Rule and analog based detection of false medical claims and so on. through such activities as analysis of ice core samples from the Antarctic and. the moisture content and the type of the soil. while minimizing the resources supplied. acquire new customers. a number of different predictive models have been proposed for future climatic conditions. (1) Analysis of telecom service purchases (2) Prediction of telephone calling patterns . like chemical fertilizers and additives (phosphate). at the same time. Few examples of data mining in the scientific environment are: (1) Studies on global climatic changes: This is a hot research area and is primarily a verification driven mining exercise. This is an example of what known as the spatial mining. (2) Studies on geophysical databases at the University of Helsinki: It has published a scientific analysis of data on farming and the environment. and develop new products. Climate data has been collected for many centuries and is being extended into the more distant past. it identified the factors affecting the crop yield. (2) Formulating Statistical Modeling of Insurance Risks (3) Using the Joint Poisson/Log-Normal Model of mining to optimize insurance policies (4) And finally finding the actuarial Credibility of the risk groups among insurers 5.5 Data Mining in Managing Scientific Data Data Mining is a concept that is taking off in the commercial sector as a means of finding useful information out of gigabytes of scientific data.Real Time Applications and Future Scope | 57 (2) Fraud detection in the voters list using neural networks in combination with symbolic and analog data mining. This can be done through: (1) evaluating the risk of the assets being insured taking into account the characteristics of the asset as well as the owner of the asset.1.7 Data Mining in Telecommunication As on this date.

so that customers can transact with all the players in supply chain.1. (3) By developing common market place services using a shared external ICT infrastructure to manage suppliers and logistics and implement electronic match making mechanisms.1.58 | Data Mining and Warehousing (3) Management of resources and network traffic (4) Automation of network management and maintenance using artificial intelligence to diagnose and repair network transmission problems. sutures. E-commerce is the newest and hottest to use data mining techniques. A recent article in the Engineering News-Record noted that e-commerce has empowered companies to collect vast amounts of data on customers—everything from the number of Web surfers in a home to the value of the cars in their garage. with the use of data mining techniques.9 Data Mining in the Retail Industries Data mining techniques have gone glove in hand in CRM (Customer Relationship Marketing) by developing models for ² predicting the propensity of customers to buy ² assessing risk in each transaction ² knowing the geographical location and distribution of customers ² analyzing customer loyalty in credit based transactions ² Assessing the competitive threat in a locality. various forms of data-mining development are being undertaken by companies looking to make sense of the raw data that has been mounting relentlessly in the recent years. This is widely used in today’s web world.8 Data Mining in Medicine and Pharmacy Widespread use of medical information systems and explosive growth of medical databases require traditional manual data analysis to be coupled with methods for efficient computer-assisted analysis. (2) By automating business interactions with customers. the article says. This was possible.) in a given surgical procedures ² Stocking the most frequently prescribed drugs And many more… 5. . 5. Few of the ways in which data mining tools find its use in e-commerce are: (1) By formulating market tactics in business operations. 5. suture materials etc.10 Data Mining in E-Commerce and the World Wide Web Sophisticated or not. Here we present some of the medical and pharmaceutical uses of Data Mining techniques for the analysis of medical databases.1. etc. ² Prediction of occurrence of disease and/or complications ² Therapy of choice for individual cancers ² Choice of antibiotics for individual infections ² Choice of a particular technique (of anastamosis.

11 Data Mining in Stock Market and Investment The rapid evolution of computer technology in the last few decades has provided investment professionals (and amateurs) with the capability to access and analyze tremendous amounts of financial data. . 5. web mining can be done in above ways. (ii) Analyzing fulfillment data to manage volume purchase contracts. In 1999 it was estimated to contain are 350 billion pages with a growth rate of 1 million page/day.12 Data Mining in Supply Chain Analysis Supply chain analysis is nothing but the analysis of the various data regarding the transactions between the supplier and the purchaser and the use of this analysis to the advantage of either of them or even both of them. (i) Helping the stock market researchers to predict future stock price movement. 5. While it may not look like the relational model.1. the hawala scandal. development of a “complete” data mining model is desirable. At present data mining is little there than a set of tools to uncover hidden information from a database. Considering the world wide web to be the largest database collection. It is of use to the supplier in the following ways: (i) It analyses the process data to manage buyer rating. (ii) Data mining of the past prices and related variables help to discover stock market anomalies like. not only will there be more efficient algorithms with better interface techniques. Data mining techniques have found wide application in supply chain analysis.2 FUTURE SCOPE The era of DBMS began using relational data model and SQL. Data archeology churns the ocean of stock market to obtain the cream of information in ways like. To develop an all-encompassing model for data mining. Coming to the purchaser side in supply chain analysis. data model and metrics for goodness. (ii) It mines payment data to advantageously update pricing policies (iii) Demand analysis and forecasting helps the supplier to determine the optimum levels of stocks and spare parts. While these are many tools to and this present. there is no all encompassing model or approach.Real Time Applications and Future Scope | 59 Determining the size of the world wide web is extremely difficult.1. As the data mining applications are of diverse types. Manual definition of request and result interpretation used on the current data mining tools may decrease with increase in automation. A major development will be in creation of a sophisticated “query language” that includes everything form SQL functions to the data mining applications. but also steps will be taken. 5. data mining techniques help them by: (i) Knowing vendor rating to choose the beneficial supplier. Over the next few years. it probably will include similar item: algorithms.

(2) CRISP-DM (Cross Industry Standard Process for Data Mining): This is a new KDD process model applicable to many different applications. Difference between these two are: TABLE 5. ² A discriminate rule: condition that is satisfied by a target class but is different from conditions satisfied in other classes ² Classification rule: set of rules used to classify data.2 Difference between SQL and DMQL SQL Allows access only to relational database Here the retrieved data is necessarily an aggregate of data from relations There is no such threshold level DMQL Allows access to background information such as concept hierarchies It indicates the type of knowledge to be mined and the retrieved data need not be a subset of data form relations It fixes a threshold that any mined information should obey A BNF statement of DMQL can be given as modified from zaigg <DMQL>: USE DATABASE (database name) {USE HIERARCHY <hierarchy name> FOR <attribute>} <rule_spec> RELATED TO <attr_or_agg_list> FROM <relation(s)> [WHERE <condition>] [ORDER BY <order_list>] {with [<kinds_of>]THRESHOLD = <threshold_value>[FOR <attribute(s)>]} The heart of a DMQL statement is the rule specification portion. The data mining request can be one of the following. The model addresses all steps in . This is where the true data mining request is made. ensure its consistency and provide concurrency and recovery features.60 | Data Mining and Warehousing A data mining query language (DMQL) based on SQL has been proposed. Two latest models that gain access via ad hoc data mining queries are: (1) KDDMS (Knowledge and Data Discover Management System): It is expected to be the future generation of data mining systems that include not only data mining tools but also techniques to manage the underlying data. ² A generalized relation: obtained by generating data from input data ² A characteristic rule: condition that is satisfied by almost all records in a target class.

k-means . k-means DB Miner Analytical System Intelligent Miner DB Miner Technologies Inc. IBM Corporation Windows Decision trees.3 DATA MINING PRODUCTS TABLE 5. neural networks (MLP. C&RT a variation of CART). k-nearest neighbour. clustering. Windows MARS Salford systems Unix (Linux. classification. logarithm) Decision trees.. OS/390. Association rules. forecasting. OS/400 JDA Intellect IDA Software Group Solaris. GRI) regression (linear. clustering Association rules. time series Association rules. including the maintenance of the results of the data mining step. classification. prediction sequential patterns. prediction. BIRCH. Sun Solaris.) . AIX. classification. Unix (linux). back propagation) regression (Linear) Naïve bayes. clustering.0. clustering factor analysis. prediction Forecasting Apriori. Windows NT Clementine SPSS Inc. RBF) Regression Windows. AIX. rule induction (C5. Decision trees (C5.. neural networks (back-probagation . (Contd. k-means clustering. sequence discovery Association rules.Real Time Applications and Future Scope | 61 the KDD process. MYS. neural networks. IBM. decision trees.0. Solaris. CARMA. Windows HP/UX. classification.3 Products of Data Mining Product 4 Thought CART Vendor Congnos Inc. Life cycle of CRISP-DM model can be described as follows: Assess (business understanding) | Access (data understanding) | Analyze (data preparation) | Act (modeling) | Automate (evaluation deployment) 5. Salford Systems Functions Prediction Classification Techniques Neural Network (MLP) Dicision tree (CART) Platforms Windows CMS . k-means.

k-means. hypothesis testing. List some of the products of data mining. time series analysis Naïve bayes ARIMA. 6. clustering. neural networks( back-propagation. prediction Classification. nonlinear. Windows Xpert Rule Miner Attar Software Ltd Association rules. RBF. classification. hierarchical clustering(agglomerative. clustering ODBC.62 | Data Mining and Warehousing Product MARS Vendor Salford Systems Functions Forecasting Techniques Regression Platforms Unix(Linux. correlation(Pearson). CHAID). regression Decision trees. Classification. SOM). logistic. exponential smoothing. regression (linear. 3. 7. Solaris). polynomial) statistical techniques (jackknife. decision trees. Windows Oracle 9i database s-plus Oracle Corporation Insightful Corporation Association rules. rules Data Miner Statsoft Inc. prediction. prediction. Windows EXERCISE 1. 2. Windows All platforms on which oracle9i runs Unix. divisive). clustering. How does data mining help the banking industry? List some of the applications of data dredging in managing scientific data. What are the uses of data mining tools in e-commerce? What is the real life application of data mining in the insurance industry? Analyse the future scopes of data harvesting in the marketing industry. MLP. 4. List a few industries which apply data mining for the management of its vast database. Monte karlo) ARIMA. . decision trees (CART. classification. 5.

we are listed some interesting examples of data storage. Kilobyte (1000 bytes) ² 1 Kilobyte: A very short story ² 2 Kilobytes: A Typewritten page ² 10 Kilobytes: An encyclopaedic page OR A deck of punched cards ² 50 Kilobytes: A compressed document image page ² 100 Kilobytes: A low-resolution photograph ² 200 Kilobytes: A box of punched cards ² 500 Kilobytes: A very heavy box of punched cards 3.1 bytes: A binary decision ² 1 byte: A single character ² 10 bytes: A single word ² 100 bytes: A telegram OR A punched card 2. It has been estimated that the amount of information in the world doubles every 20 months and the sizes as well as number of databases are increasing even faster.1 THE CALCULATIONS FOR MEMORY CAPACITY The past couple of decades have seen a dramatic increase in the amount of information or data being stored in electronic format.CHAPTER 6 DATA WAREHOUSE EVALUATION 6. Bytes (8 bits) ² 0. financial data in banking and securities.5 inch floppy disk ² 2 Megabytes: A high resolution photograph . policy and claim data in insurance. Point of sale data in retail. Here. There are many examples that can be cited. medical history data in health care. 1. This accumulation of data has taken place at an explosive rate. are some instances of the types of data that is being collected. Megabyte (1 000 000 bytes) ² 1 Megabyte: A small novel OR A 3.

64 | Data Mining and Warehousing ² 5 Megabytes: The complete works of Shakespeare OR 30 seconds of TV-quality video ² 10 Megabytes: A minute of high-fidelity sound OR A digital chest X-ray ² 20 Megabytes: A box of floppy disks ² 50 Megabytes: A digital mammogram ² 100 Megabytes: 1 meter of shelved books OR A two-volume encyclopaedic book ² 200 Megabytes: A reel of 9-track tape OR An IBM 3480 cartridge tape ² 500 Megabytes: A CD-ROM 4. Yottabyte (1 000 000 000 000 000 000 000 000 bytes) . 8. Gigabyte (1 000 000 000 bytes) ² 1 Gigabyte: A pickup truck filled with paper OR A symphony in high-fidelity sound OR A movie at TV quality ² 2 Gigabytes: 20 meters of shelved books OR A stack of 9-track tapes ² 5 Gigabytes: An 8mm Exabyte tape ² 10 Gigabytes: ² 20 Gigabytes: A good collection of the works of Beethoven ² 50 Gigabytes: A floor of books OR Hundreds of 9-track tapes ² 100 Gigabytes: A floor of academic journals OR A large ID-1 digital tape ² 200 Gigabytes: 50 Exabyte tapes 5. Petabyte (1 000 000 000 000 000 bytes) ² ² ² ² 1 Petabyte: 3 years of EOS data (2001) 2 Petabytes: All US academic research libraries 20 Petabytes: Production of hard-disk drives in 1995 200 Petabytes: All printed material OR Production of digital magnetic tape in 1995 7. Zettabyte (1 000 000 000 000 000 000 000 bytes) 9. Exabyte (1 000 000 000 000 000 000 bytes) ² 5 Exabytes: All words ever spoken by human beings. Terabyte (1 000 000 000 000 bytes) ² 1 Terabyte: An automated tape robot OR All the X-ray films in a large technological hospital OR 50000 trees made into paper and printed OR Daily rate of EOS data (1998) ² 2 Terabytes: An academic research library OR A cabinet full of Exabyte tapes ² 10 Terabytes: The printed collection of the US Library of Congress ² 50 Terabytes: The contents of a large Mass Storage System 6.

or signs knowledge. which is called pragmatics.e.e. “There is. i. . is introduced. defines an important characteristic of knowledge. their semantics adds a second dimension to signs. (2) The relation between signs and their meaning. can be called one-dimensional. Normalization in a relational database is the process of removing redundancy in data by separating the data into multiple tables. A table in a database looks like a simple spreadsheet.1 Provides Example of Employee table with 2 columns and 3 rows.. The example above has 3 rows and 2 columns in employee table. INFORMATION AND KNOWLEDGE In this section we will present three terms widely—but often unreflectingly—used in several ITrelated (and other) communities.” (1) The relation among signs.. i.. signs only viewed under this dimension equal data.e.1 Example for Database Emp. i. The definitions will be oriented according to the three dimensions of semiotics (the theory of signs). for example the column Emp_Number. If this third dimension including users. and pragmatics. syntax. (3) The third dimension is added by relating signs and their ‘users’. A row (= tuple. in general. we call the patterns. Information is a meaningful data and Knowledge is a interesting and useful patterns in a data base. entry) is a group of related data. number 101 102 103 Name John Jill Jack A table is a matrix with data. i. TABLE 6. Table 6. One column (data element) contains data of one and the same kind. the syntax of ‘sentences’ does not relate signs to anything in the real world and thus. Data means any symbolic reperesentation of facts or ideas from which information can potentially be extracted.1 Definitions A database is a collection of tables. who pursue goals which require performing actions. ‘information’. In our eyes. that data becomes information. 6.e. semantics.Data Warehouse Evaluation | 65 6. with related data... i.3. no known way to distinguish knowledge from information or data on a purely representational basis. This relationship. and ‘knowledge’. performed by actors. for example the data of one employee detail. ‘data’. only knowledge enables actions and decisions.e.3 FUNDAMENTAL OF DATABASE 6. It is only through this relation between sign and meaning.2 DATA.

The transformation consists of the following: ² Transforming entities into tables ² Transforming attributes into columns ² Transforming domains into data types and constraints To support the mapping of attributes to table columns you will need to map each logical domain of the attribute to a physical data type and perhaps additional constraints. ² Details regarding the manner in which the DBMS supports indexing. In a physical database.3. But no commercial DBMS product supports relational domains. each column must be assigned a data type. A logical data model is required before you can even begin to design a physical database. referential integrity. such as graphic. Assuming that the logical data model is complete. Therefore the domain assigned in the logical data model must be mapped to a data type supported by the DBMS. 6. Certain data types require a maximum length to be specified. For example. you can create an effective and efficient database from a logical data model. For example a character data type could be specified as VARCHAR(10). To successfully create a physical database design you will need to have a good working knowledge of the features of the DBMS including: ² In-depth knowledge of the database objects supported by the DBMS and the physical structures and files required to support those objects. indicating that up to 10 characters can be stored for the column. constraints. data types.2 Data Base Design Database design is the process of transforming a logical data model into an actual physical database. The first step in transforming a logical data model into a physical model is to perform a simple translation from logical terms to physical objects. floating point. ² Knowledge of the DBMS configuration parameters that are in place. and decimal (which require a length and scale) types. ² Data definition language (DDL) skills to translate the physical design into actual database objects. this simple transformation will not result in a complete and correct physical database design – it is simply the first step. though. ² Detailed knowledge of new and obsolete features for particular versions or releases of the DBMS to be used. what must be done to implement a physical database? The first step is to create an initial physical data model by transforming the logical data model into a physical implementation based on an understanding of the DBMS to be used for deployment.66 | Data Mining and Warehousing The process of removing redundancy in data by separating the data into multiple tables. You may need to apply a length to other data types as well. Of course. and other features that augment the functionality of database objects. what data type and length will be used for monetary values if no built-in currency data type exists? Many of the major DBMS . Armed with the correct information. You may need to adjust the data type based on the DBMS you use.

As a first course of action you should try to use the primary key as selected in the logical data model. 6. ² For character columns. there are many other decisions that must be made during the transition from logical to physical. reverse key. 1 through 10. and so on.g. Of course. Without a constraint. For example.3. However. etc. data files.Data Warehouse Evaluation | 67 products support user-defined data types. if no built-in data type is acceptable.. zero. A constraint must be added to restrict the values that can be stored for the column to the specified range. Consider a domain of integers between 1 and 10 inclusive.2 Packages—Example RDBMS name Oracle IBM DB2 UDB IBM Informix Microsoft SQL Server Sybase Terradata Company name Oracle Corporation IBM Corporation IBM Corporation Microsoft Sybase Corporation NCR . multiple candidate keys often are uncovered during the data modeling process. You may decide to choose a primary key other than the one selected during logical design – either one of the candidate keys or another surrogate key for physical implementation. should fixed length or variable length be used? ² Should the DBMS be used to assign values to sequences or identity columns? ² Implementing logical relationships by assigning referential constraints. hash. de normalization. A primary key should be assigned for every entity in the logical data model. you also may need to apply a constraint to the column. ² Building indexes on columns to improve query performance. buffer pool specification. bit map. partitioning. so you might want to consider creating a data type to support the logical domain. ² Other physical aspects such as column ordering.3 Selection for the Database/Hardware Platform RDBMS are used in OLTP applications (e. ² Deciding on the clustering sequence for the data. and values greater than ten could be stored. negative numbers. TABLE 6. But even if the DBMS does not mandate a primary key for each table it is a good practice to identify a primary key for each physical table you create. ² Choosing the type of index to create: b-tree. In addition to a data type and length. Specification of a primary key is an integral part of the physical design of entities and attributes. ATM cards) very frequently and sometimes data warehouse may also use relational databases. Using check constraints you can place limits on the data values that can be stored in a column or set of columns. Popular RDBMS Databases are listed below. Simply assigning the physical column to an integer data type is insufficient to match the domain. Failure to do so will make processing the data in that table more difficult. each of the following must be addressed: ² The nullability (NULL) of each column in each table.

A multi-dimensional structure is arranged so that every data item is located and accessed based on the intersection of the dimension members . as well as through analysis of historical and projected data in various “what-if” data model that it would be the best way for any large computations to be done within 5 years. Parallel Processing Support: The days of multi-million dollar supercomputers with one single CPU are gone. bugs and bug fixes are often hardware dependent. 6. This is achieved through use of an OLAP server. OLAP functionality is characterized by dynamic multi-dimensional analysis of consolidated enterprise data supporting end user analytical and navigational activities including: ² calculations and modeling applied across dimensions. As a result. regardless of database size and complexity. consistent. through hierarchies and/or across members ² trend analysis over sequential time periods ² slicing subsets for on-screen viewing ² drill-down to deeper levels of consolidation ² reach-through to underlying detail data ² rotation to new dimensional comparisons in the viewing area OLAP is implemented in a multi-user client/server mode and offers consistently rapid response to queries. personalized viewing. where each processor can perform a part of the task. OLAP helps the user synthesize enterprise information through comparative.68 | Data Mining and Warehousing In making selection for the database/hardware platform. one needs to determine the approximate amount of data that is to be kept in the data warehouse system once it’s mature.1 OLAP Definition On-Line Analytical Processing (OLAP) is a category of software technology that enables analysts. and nowadays the most powerful computers all use multiple CPUs.4. managers and executives to gain insight into data through fast.4 OLAP AND OLAP SERVER 6. interactive access to a wide variety of possible views of information that has been transformed from raw data to reflect the real dimensionality of the enterprise as understood by the user. and base any testing numbers from there. Massively parallel computers in 1993.2 OLAP Server An OLAP server is a high-capacity. all at the same time. 6. there are going to be certain parts of the code that is hardware platform-dependent. multi-user data manipulation engine specifically designed to support and operate on multi-dimensional data structures. RDBMS/Hardware Combination: Because the RDBMS physically sits on the hardware platform.4. there are several items that need to be carefully considered: Scalability: How can the system grow as your data storage needs grow? Which RDBMS and hardware platform can handle large sets of data most efficiently? To get an idea of this.

If data is allowed to accumulate in the OLTP so it can be used for analysis. valid data. flexible calculation and transformation of raw data based on formulaic relationships. or allowed to accumulate in the OLTP database. If data is simply archived for preservation. Because the purpose of a data warehouse differs from that of an On Line Transaction Processing (OLTP). The large indexes needed to support these queries . 6. or it may populate its data structures in real-time from relational or other databases. unpredictable queries that access many rows per table Loaded with consistent. as well as for fast. These queries access and process large portions of the continually growing historical data and add a substantial load to the database.3 Comparison of Data Warehouse Database and OLTP Data warehouse database Designed for analysis of business measures by categories and attributes Optimized for bulk loads and large. and by providing services that would complicate and degrade OLTP operations if they were performed in the OLTP database. staging the multi-dimensional data in the OLAP server is often the preferred method.Data Warehouse Evaluation | 69 which define that item. Given the current state of technology and the end user requirement for consistent and rapid response times. usually adding or retrieving a single row at a time per table Optimized for validation of incoming data during transactions. The OLAP server may either physically stage the processed multi-dimensional information to deliver consistent and rapid response times to end users. OLAP AND DATA MINING A relational database is designed for a specific purpose. The design of the server and the structure of the data are optimized for rapid ad-hoc information retrieval in any orientation. complex.1 A Data Warehouse Supports OLTP A data warehouse supports an OLTP system by providing a place for the OLTP database to offload data as it accumulates. it is not available or organized for use by analysts and decision makers. data is archived to static media such as magnetic tape. the OLTP database continues to grow in size and requires more indexes to service analytical and report queries. TABLE 6.5. or offer a choice of both.5 DATA WAREHOUSES. Without a data warehouse to hold historical information. requires no real time validation Supports few concurrent users relative to OLTP OLTP database Designed for real-time business operations Optimized for a common set of transactions. the design characteristics of a relational database that supports a data warehouse differ from the design characteristics of an OLTP database. An OLTP system is designed to handle a large volume of small-predefined transactions. OLTP. uses validation data tables Supports thousands of concurrent users 6.

OLAP is not designed to store large volumes of text or binary data. 6.5. as well as data in the relational data warehouse database. and then preprocesses these cubes to provide maximum performance for queries that summarize data in various ways. . High volume analytical and reporting queries are handled by the data warehouse and do not load the OLTP. OLAP organizes data warehouse data into multidimensional cubes based on this dimensional model. Analysis Services provides tools for developing OLAP applications and a server specifically designed to service OLAP queries. data mining results can be incorporated into OLAP cubes to further enhance model-driven analysis by providing an additional dimensional viewpoint into the OLAP model. In SQL server 2000. In addition. Thus. business objects and data stage. Data mining has traditionally operated only on raw data in the data warehouse database or. In SQL server 2000. As data is moved to the data warehouse. it is also reorganized and consolidated so that analytical queries are simpler and more efficient. Whereas OLAP organizes data in a model suited for exploration by analysts. data mining performs analysis on data and provides the results to decision makers. Other tolls are informatica. more commonly. These queries can also be complicated to develop due to the typically complex OLTP database schema.3 Data Mining is a Data Warehouse Tool Data mining is a technology that applies sophisticated and complex algorithms to analyze data and expose interesting information for analysis by decision makers. data mining can be used to analyze sales data against customer attributes and create a new cube dimension to assist the analyst in the discovery of the information embedded in the cube data. text files of data extracted from the data warehouse database. The inherent stability and consistency of historical data in a data warehouse enables OLAP to provide its remarkable performance in rapidly summarizing information for analytical queries. a query that requests the total sales income and quantity sold for a range of products in a specific geographical region for a specific time period can typically be answered in a few seconds or less regardless of how many hundreds of millions of rows of data are stored in the data warehouse database. which does not need additional indexes for their support. For example. 6. A data warehouse offloads the historical data from the OLTP.2 OLAP is a Data Warehouse Tool Online analytical processing (OLAP) is a technology designed to provide superior performance for ad hoc business intelligence queries. nor is it designed to support high volume update transactions. For example.70 | Data Mining and Warehousing also tax the OLTP transactions with additional index maintenance.5. A data warehouse provides a multidimensional view of data in an intuitive model designed to match the types of queries posed by analysts and decision makers. OLAP is designed to operate efficiently with data organized in accordance with the common dimensional model used in data warehouses. OLAP supports model-driven analysis and data mining supports data-driven analysis. allowing the OLTP to operate at peak transaction efficiency. analysis services provides data mining technology that can analyze data in OLAP cubes.

5. multidimensional integrated consolidated ad-hoc lots of scans complex query millions hundreds 100GB–TB query throughout.5. 2. 3.4 Difference between OLTP and OLAP TABLE 6.Data Warehouse Evaluation | 71 6. What is the difference between data and information? What are OLAP and OLAP Server? What is OLTP? What are the items to be considered to create a data warehouse? Compare OLAP and OLTP. Distinguish between data base warehouse and OLTP data base. up-to-date detailed. 6.4 Difference between OLTP and OLAP OLTP OLAP knowledge worker decision support subject-oriented historical. simple transaction tens thousands 100MB–GB transaction throughout usage access unit of work # records accessed # users DB size metric EXERCISE 1. . flat relational isolated repetitive read/write index/hash on prim. key short. 4. response users function DB design data clerk. summarized. IT professional day to day operations application-oriented current.

“normalized”. A very frequent problem in enterprises is the impossibility for accessing to corporate. quality improvement and integration of data that comes from operational bases. which is a data base organized according to the corporate data model. the T-shirt is in another. it is the result of transformations. and the CDs are in a third. it includes indicators that are derived from operational data and give it additional value. complete and integrated information of the enterprise that can satisfy decision-making requests.CHAPTER 7 DATA WAREHOUSE DESIGN 7. We found in the literature. while its maintenance does not suppose transactional load. . 7. This is accomplished by organizing it according to the enterprise’s corporate data model. and another that bases mainly in the concept of materialized views.1 INTRODUCTION Data Warehouse (DW) is a database that stores information oriented to satisfy decision-making requests. Concerning the data it contains. in a DW environment end users make queries directly against the DW through user-friendly query tools. A paradox occurs: data exists but information cannot be obtained. it is supposed to support complex queries (summarization. crossing of data). A data warehouse has three main components: ² A “Central Data Warehouse” or “Operational Data Store(ODS)”. ² The “legacy systems” where an enterprise’s data are currently kept. A DW is a database with particular features. In general.2 THE CENTRAL DATA WAREHOUSE The Central Data Warehouse is just that—a warehouse. aggregates. globally two different approaches for Relational DW design: one that applies dimensional modeling techniques. In this chapter we concentrate in DW design. Besides. Building and maintaining a DW need to solve problems of many different aspects. Think of it as a giant grocery store warehouse where the chocolates are kept in one section. in order to minimize redundancy and so that each may be found easily. In addition. a DW is constructed with the goal of storing and providing all the relevant information that is generated along the different databases of an enterprise. All the enterprise’s data are stored in there. instead of accessing information through reports generated by specialists. Concerning its utilization. ² One or more “data marts”—extracts from the central data warehouse that are organized according to the particular retrieval requirements of individual users.

country Fig. date 3-D (base) cuboid Product. The basic concepts of dimensional modeling are: facts. Data redundancy is not a problem in DWs because data is not updated on-line. According to the objectives of dimensional modeling are: (i) To produce database structures that are easy for end-users to understand and write queries against. usually end-users interact with the database through a layer of software. but not for DWs: (i) Its structure is not easy for end-users to understand and use. but tends to penalize retrievals. country 2-D cuboids Product. 0-D (apex) cuboid Product Date try t.1 Cubes . This maximizes efficiency of updates. OLAP cube design requirements will be a natural outcome of the dimensional model if the data warehouse is designed to support the way users wants to query data. 7. A measure is a numeric attribute of a fact. A fact is a collection of related data items. Normalized databases have some characteristics that are appropriate for OLTP systems. Dimensions determine the contextual background for the facts. It typically represents business items or business transactions. In OLTP systems this is not a problem because. they are the parameters over which we want to perform OLAP. consisting of measures and context data. (ii) To maximize the efficiency of queries. dimensions and measures. A dimension is a collection of data that describe one business dimension.Data Warehouse Design | 73 Dimensional models represent data with a “cube” structure. representing the performance or behavior of the business relative to the dimensions. coun Produc Country 1-D cuboids Date. (ii) Data redundancy is minimized. making more compatible logical data representation with OLAP data management. It achieves these objectives by minimizing the number of tables and relationships between them. date.

the fact table can be reduced to columns for dimension foreign keys and numeric fact values. Location.3. usually numeric and additive. Examples include sales. Text. Fact tables are often very large. Examples are customers. Though most facts are additive.1 Fact Tables A fact table typically has two types of columns: those that contain numeric facts (often called measurements). Dimension tables are usually textual and descriptive and you can use them as the row headers of the result set. The definitions of this ‘sales’ fact table follow: . Non-additive facts cannot be added at all. A fact table usually contains facts with the same level of aggregation. where you cannot tell what a level means simply by looking at it. that can be analyzed and examined. Time. Semi-additive facts can be aggregated along some of the dimensions and not along others. and profit. Fact tables contain business event details for summarization. Dimension tables store the information you normally use to contain queries. Because dimension tables contain records that describe facts.74 | Data Mining and Warehousing 7. containing hundreds of millions of rows and consuming hundreds of gigabytes or multiple terabytes of storage. 7. and denormalized data are typically not stored in the fact table. they can also be semi-additive or non-additive. contain the relatively static data in the warehouse. 7. cost. Fact tables represent data. Fact tables typically contain facts and foreign keys to the dimension tables. and those that are foreign keys to dimension tables. A fact table contains either detail-level facts or facts that have been aggregated. From a modeling standpoint. the primary key of the fact table is usually a composite key that is made up of all of its foreign keys. Creating a new fact table: You must define a fact table for each star schema. Additive facts can be aggregated by simple arithmetical addition. Time Customer Sales Facts Seller Product Manufacturing Location Fig. Fact tables that contain aggregated facts are often called SUMMARY TABLES.2 A sales analysis star schema Dimension tables. BLOBs. An example of this is averages. An example of this is inventory levels. Suppliers or products. also known as lookup or reference tables.3 DATA WAREHOUSING OBJECTS The following types of objects are commonly used in dimensional data warehouse schemas: Fact tables are the large tables in your warehouse schema that store business measurements. A common example of this is sales.

and time. The definitions of this ‘customer’ fact table follow: .2 Dimension Tables A dimension is a structure. textual values. products. Partitioned fact tables can be viewed as one table with an SQL UNION query as long as the number of tables involved does not exceed the limit for a single query. These natural rollups or aggregations within a dimension table are called hierarchies. inventory. several conforming dimension tables. Each business function should have its own fact table and will probably have some unique dimension tables. and finance. OLAP cubes are usually partitioned to match the partitioned fact table segments for ease of maintenance. For example. and some dimension tables unique to the specific business function. Any dimensions that are common across the business functions must represent the dimension information in the same way. They are normally descriptive. Such business-specific schemas may be part of the central data warehouse or implemented as data marts. If fact tables are partitioned. 7. enable you to answer business questions. Commonly used dimensions are customers. time_id DATE CONSTRAINT sales_time_nn NOT NULL.” Each business function will typically have its own schema that contains a fact table.2) CONSTRAINT sales_amount_nn NOT NULL. cust_id NUMBER CONSTRAINT sales_customer_nn NOT NULL.Data Warehouse Design | 75 CREATE TABLE sales ( prod_id NUMBER(7) CONSTRAINT sales_product_nn NOT NULL. ad_id NUMBER(7). and also in one or more departmental data marts.3. cost NUMBER(10. Using conforming dimensions is critical to successful data warehouse design. and the time dimension is the most common one to use because of the historical nature of most data warehouse data. or product that is used in multiple schemas is called a conforming dimension if all copies of the dimension are the same. A dimension such as customer. Summarization data and reports will not correspond if different schemas use different versions of a dimension table. The partition divisions are almost always along a single dimension. combined with facts. Dimension data is typically collected at the lowest level of detail and then aggregated into higher-level totals that are more useful for analysis. time.2) CONSTRAINT sales_cost_nn NOT NULL ) Multiple Fact Tables: Multiple fact tables are used in data warehouses that address multiple business functions. such as sales. as discussed earlier in “Dimension Tables. amount NUMBER(10. often composed of one or more hierarchies. A dimension table may be used in multiple places if the data warehouse contains multiple fact tables or contributes data to data marts. a product dimension may be used with a sales fact table and an inventory fact table in the data warehouse. that categorizes data. Very large fact tables may be physically partitioned for implementation and maintenance design considerations. Several distinct dimensions. Dimensional attributes help to describe the dimensional value.quantity_sold NUMBER(4) CONSTRAINT sales_quantity_nn NOT NULL.

there may be a number of sales to a single customer.country_id CHAR(2) CONSTRAINT customer_country_id_nn NOT NULL. Year. Attributes serve as report labels and query constraints. cust_first_name VARCHAR2(20) CONSTRAINT customer_fname_nn NOT NULL.prod_subcategory) LEVEL category IS (products. A dimension may contain multiple hierarchies – a time dimension often contains both calendar and fiscal year hierarchies. or a number of sales of a single product.prod_subcat_desc ATTRIBUTE category DETERMINES products. cust_last_name VARCHAR2(40) CONSTRAINT customer_lname_nn NOT NULL.76 | Data Mining and Warehousing CREATE TABLE customers ( cust_id NUMBER. The records in a dimension table establish one-to-many relationships with the fact table.cust_city VARCHAR2(30) CONSTRAINT customer_city_nn NOT NULL. product category may exist as a simple integer in the OLTP database.prod_desc ATTRIBUTE subcategory DETERMINES products. Hierarchies are determined by the business need to group and summarize data into usable information. Quarter. cust_credit_limit NUMBER. cust_state_district VARCHAR2(40). cust_postal_code VARCHAR2(10) CONSTRAINT customer_pcode_nn NOT NULL.prod_cat_desc. Hierarchies: The data in a dimension is usually hierarchical in nature.cust_street_address VARCHAR2(40) CONSTRAINT customer_st_addr_nn NOT NULL.prod_category) HIERARCHY prod_rollup ( product CHILD OF subcategory CHILD OF category ) ATTRIBUTE product DETERMINES products. The dimension table contains attributes associated with the dimension entry. but the dimension table should contain the actual text for the category. it is usually a hierarchy that . these attributes are rich and user-oriented textual details. The code may also be carried in the dimension table if needed for maintenance. Attributes that are coded in an OLTP database should be decoded into descriptions. cust_email VARCHAR2(30) ) CREATE DIMENSION products_dim LEVEL product IS (products. cust_phone_number VARCHAR2(25). However. For example.prod_id) LEVEL subcategory IS (products. cust_income_level VARCHAR2(30). For example.cust_sex CHAR(1). For example. a time dimension often contains the hierarchy elements: (all time). Month. Geography is seldom a dimension of its own. if a dimension attribute changes frequently. maintenance may be easier if the attribute is assigned to its own table to create a snowflake dimension. cust_marital_status VARCHAR2(20).prod_name ATTRIBUTE product DETERMINES products. such as product name or customer name and address. cust_year_of_birth NUMBER(4). Day or Week. This denormalization simplifies and improves the efficiency of queries and simplifies user query tools.

region. customers. If the number of discrete values creates a very large table of all possible value combinations. unrelated dimensions into a single physical dimension. often called a junk dimension. collecting these comments in a single dimension removes a sparse text field from the fact table and replaces it with a compact foreign key. inventory.Data Warehouse Design | 77 imposes a structure on sales points. An example geography hierarchy for sales points is: (all). the table can be populated with value combinations as they are encountered during the load or update process. 7. This can greatly reduce the size of the fact table by reducing the number of foreign keys in fact table records. For example. However. Another multiuse dimension might contain useful textual comments that occur infrequently in the source data records. A data warehouse must be designed to satisfy the following requirements: ² Deliver a great user experience—user acceptance is the measure of success ² Function without interfering with OLTP systems ² Provide a central repository of consistent data ² Answer complex queries quickly ² Provide a variety of powerful analytical tools such as OLAP and data mining Most successful data warehouses that meet these requirements have these common characteristics: ² Based on a dimensional model ² Contain historical data ² Include both detailed and summarized data ² Consolidate disparate data from multiple sources while retaining consistency ² Focus on a single subject such as sales. city. A common example of a multi-use dimension is a dimension that contains customer demographics selected for reporting standardization. Hierarchies are also essential components in enabling more complex rewrites. or finance Data warehouses are often quite large. state or district. Often the combined dimension will be prepopulated with the cartesian product of all dimension values. . the database can aggregate existing sales revenue on a quarterly base to a yearly aggregation when the dimensional dependencies between quarter and year are known.4 GOALS OF DATA WAREHOUSE ARCHITECTURE A data warehouse exists to serve its users—analysts and decision makers. country. Multi-use dimensions: Sometimes data warehouse design can be simplified by combining a number of small. They define the parent-child relationship between the levels in a hierarchy. size is not an architectural goal—it is a characteristic driven by the amount of data needed to serve the users. store. or other geographically distributed dimensions. Level relationships specify top-to-bottom ordering of levels from most general (the root) to most specific information.

After a few iterations.5 DATA WAREHOUSE USERS The success of a data warehouse is measured solely by its acceptance by users. Their work can contribute to closed loop systems that deeply influence the operations and profitability of the company. they will probably never compose a true ad hoc query. They will figure out how to quantify a subject area. information consumers. Information Consumers: Most users of the data warehouse are Information Consumers. They use static or simple interactive reports . historical data might as well be archived to magnetic tape and stored in the basement. knowledge workers. Data warehouse users can be divided into four categories: Statisticians. Knowledge Workers are often deeply engaged with the data warehouse design and place the greatest demands on the ongoing data warehouse operations team for training and support. Knowledge Workers: A relatively small number of analysts perform the bulk of new queries and analyses against the data warehouse. 7. Successful data warehouse design starts with understanding the users and their needs.78 | Data Mining and Warehousing 7. and executives.3 Data warehouse users Statisticians: There are typically only a handful of statisticians and operations research types in any organization. These are the users who get the Designer or Analyst versions of user access tools. Statisticians (2%) Knowledge Workers (15%) Executives Information Consumers (82%) Fig. their queries and reports typically get published for the benefit of the Information Consumers. Without users. Each type makes up a portion of the user population as illustrated in this diagram.

5. Direct queries to the data warehouse relational database should be limited to those that cannot be accomplished through existing tools.6. Integrity constraints provide a mechanism for ensuring that data conforms to guidelines specified by the database administrator. and published reports are highly visible. 7. the star or snowflake schema is created in the relational database. The composite primary key in the fact table is an expensive key to maintain. and gather feedback from these users to improve the information sites over time.1 Keys and Relationships Tables are implemented in the relational database after surrogate keys for dimension tables have been defined and primary and foreign keys and their relationships have been identified. Information consumers do not interact directly with the relational database but may receive e-mail reports or access web pages that expose data from the relational database. and fact table partitions are also defined. and Information Consumers will use interactive reports designed by others. When using the analysis services tools in SQL server 2000. Primary/ foreign key relationships should be established in the database schema. analysts will write MDX queries against OLAP cubes and use data mining. Executives use standard reports or ask others to create specialized reports for them. statisticians will often perform data mining. The index alone is almost as large as the fact table. Analysts may write complex queries to extract and compile specific information not readily accessible through existing tools. In many scenarios a clustered primary key provides excellent . 7. Statisticians frequently extract data for use by special analytical tools.Data Warehouse Design | 79 that others have developed. which are often more efficient than direct queries and impose less load on the relational database. OLAP cubes are designed that support the needs of the users. Views. Reporting tools and custom applications often access the database directly.. surrogate keys are defined and primary and foreign key relationships are established. 7. Executives: Executives are a special case of the Information Consumers group. Set up a great communication infrastructure for distributing information widely. The most common types of constraints include: ² UNIQUE constraints—To ensure that a given column is unique ² NOT NULL constraints—To ensure that no null values are allowed ² FOREIGN KEY constraints—To ensure that two keys share a primary key to foreign key relationship Constraints can be used for these purposes in a data warehouse is Data cleanliness and Query optimization. They usually interact with the data warehouse only through the work product of others.1 How Users Query the Data Warehouse? Information for users can be extracted from the data warehouse relational database or from the output of analytical services such as OLAP or data mining.6 DESIGN THE RELATIONAL DATABASE AND OLAP CUBES In this phase. The index on the primary key is often created as a clustered index. This group includes a large number of people. indexes.

and recommends an index configuration to improve the performance of the database. many star schemas are defined with an integer surrogate primary key. distribution of queries among tables. or executed manually at a later time. All indexes on the fact table will be large. With a large number of dimensions. or the internals of SQL Server. The SQL Server 2000 Index Tuning Wizard allows you to select and create an optimal set of indexes and statistics for a database without requiring an expert understanding of the structure of the database. it can then be implemented immediately. ² It analyzes the effects of the proposed changes. it is usually more effective to create a unique clustered index on a meaningless IDENTITY column. The wizard analyzes a query workload captured in a SQL Profiler trace or provided by an SQL script.2 Indexes Dimension tables must be indexed on their primary keys. A recommendation from the wizard consists of SQL statements that can be executed to create new. such as disk space constraints. if wanted. should the database administrator determine this structure would provide better performance. Indexed views are recommended on platforms that support their use.6. Also create an IDENTITY column in the fact table that could be used as a unique clustered index. As a result. 7. ² It allows you to customize its recommendations by specifying advanced options. We recommend that the fact table be defined using the composite primary key. scheduled as a SQL Server job.80 | Data Mining and Warehousing query performance. The larger the number of dimensions in the schema. The index tuning wizard provides the following features and functionality: ² It can use the query optimizer to analyze the queries in the provided workload and recommend the best combination of indexes to support the query mix in the workload. ² It can recommend ways to tune the database for a small set of problem queries. or no primary key at all. the less beneficial it is to cluster the primary key index. The fact table must have a unique index on the primary key. . and then employed periodically during ongoing operation to maintain indexes in tune with the user query workload. more effective indexes and. drop existing indexes that are ineffective. However. the system will require significant additional storage space. After the Index Tuning Wizard has suggested a recommendation. The empirical tuning approach provided by the index tuning wizard can be used frequently when the data warehouse is first implemented to develop the initial index set. which has sophisticated index techniques and an easy to use Index Tuning Wizard tool to tune indexes to the query workload. Elaborate initial design and development of index plans for end-user queries is not necessary with SQL server 2000. There are scenarios where the primary key index should be clustered and other scenarios where it should not. all other indexes on the fact table use the large clustered index key. the workload. and performance of queries in the workload. including index usage. which are the surrogate keys created for the data warehouse tables. and query performance may degrade.

Indexed views can be used to improve performance of user queries that access data through views.1. a single object (the fact table) sits in the middle and is radically connected to other surrounding objects (dimension lookup tables) like a star. subregion name).1 Dimensional Model Schemas The principal characteristic of a dimensional model is a set of detailed business facts surrounded by multiple dimensions that describe those facts. 7. The following diagram shows a classic star schema. In the star schema design. Users can be granted access to views without having access to the underlying data. the schema for a dimensional model contains a central fact table and multiple dimension tables.7. location dimension. 7. indexes. branch name. ² Identify dimensions for facts (product dimension. When realized in a database.7 DATA WAREHOUSING SCHEMAS A schema is a collection of database objects. ² List the columns that describe each dimension (region name. storage capacity. . 7. The physical implementation of the logical data warehouse model may require some changes to adapt it to your system parameters—size of machine. A star schema can be simple or complex. organization dimension).Data Warehouse Design | 81 7. Steps in Designing Star Schema ² Identify a business process for analysis (like sales). View definitions should create column and table names that will make sense to business users. including tables. The model of your source data and the requirements of your users help you design the data warehouse schema. and software. A simple star consists of one fact table. Most data warehouses use a dimensional model. ² Identify measures or facts (sales dollar).3 Views Views should be created for users who need direct access to data in the data warehouse relational database.1 Star Schemas A schema is called a star schema if all dimension tables can be joined directly to the fact table. ² Determine the lowest level of summary in a fact table (sales dollar).7.6. a complex star can have more than one fact table. and synonyms. A dimensional model may produce a star schema or a snowflake schema. You can sometimes get the source model from your company’s enterprise data model and reverse-engineer the logical data model for the data warehouse from this. number of users. views. You can arrange schema objects in the schema models designed for data warehousing in a variety of ways. time dimension. type of network.

for example.82 | Data Mining and Warehousing Products Dimension Table Sales (amount sold. . 7. quantity sold) Fact Table Customers Dimension Table Fig. 7. in a time dimension. regardless of whether the levels in the hierarchy represent aggregated totals or not.5 Example of star schema Hierarchy A logical structure that uses ordered levels as a means of organizing data.4 Star schema Times Dimension Table Channels Dimension Table Example – Star Schema with Time Dimension PRODUCT DIMENSION Product Dimension Identifier (PK) Product Category Name Product Sub-Category Name Product Name Product Feature Description Date Time Stamp ORGANISATION DIMENSION Organisation Dimension Identifier (PK) Corporate Office Name Region Name Branch Name Employees Name Date Time Stamp SALES FACT Time Dimension Identifier (FK) Location Dimension Identifier (FK) Product Dimension Identifier (FK) Organisation Dimension Identifier (FK) LOCATION DIMENSION Location Dimension Identifier (PK) Country Name State Name City Name Date Time Stamp Sales Dollar Date Time Stamp TIME DIMENSION Time Dimension Identifier (PK) Year Number Day of Year Quarter Number Month Number Month Name Month Day Number Week Number Day of Week Calender Date Date Time Stamp Fig. A hierarchy can be used to define data aggregation. from the quarter level to the year level. A hierarchy can also be used to define a navigational drill path. a hierarchy might be used to aggregate data from the month level to the quarter level.

a time dimension might have a hierarchy that represents data at the month. Suppliers Times Sales (amount sold. product.7. quarter.6 A snowflake schema . quantity sold) Customers Channels Products Countries Fig. For example. Fact Table A table in a star schema that contains facts and connected to dimensions. ² Sales dollar value for a particular product ² Sales dollar value for a product in a location ² Sales dollar value for a product in a year within a location ² Sales dollar value for a product in a year within a location sold or serviced by an employee 7. and year levels. A fact table usually contains facts with the same level of aggregation. For example. A fact table typically has two types of columns: those that contain facts and those that are foreign keys to dimension tables.Data Warehouse Design | 83 Level A position in a hierarchy. 7. It shows that data can be sliced across all dimensions and again it is possible for the data to be aggregated across multiple dimensions.1. sales fact table is connected to dimensions location. “Sales Dollar” in sales fact table can be calculated across all dimensions independently or in a combined manner which is explained below. In the example Fig.6. A fact table might contain either detail level facts or facts that have been aggregated (fact tables that contain aggregated facts are often instead called summary tables). a dimension that describes products may be separated into three tables (snowflaked). 7. time and organization. The primary key of a fact table is usually a composite key that is made up of all of its foreign keys.2 Snowflake Schemas A schema is called a snowflake schema if one or more dimension tables do not join directly to the fact table but must join through other dimension tables.

The reason is that hierarchies(category. and month) are being broken out of the dimension tables (PRODUCT. state. The main advantage of the snowflake schema is the improvement in query performance due to minimized disk storage requirements and joining smaller lookup tables. 7. LOCATION.84 | Data Mining and Warehousing The snowflake schema is an extension of the star schema where each point of the star explodes into more points. and TIME) PRODUCT CATEGORY LOOKUP Product Category Code (PK) Product Category Name Date Time Stamp BRANCH LOOKUP Branch Code (PK) Branch Name Date Time Stamp PRODUCT DIMENSION Product Dimension Identifier (PK) Product Category Code (FK) Product Sub-Category Name Product Name Product Feature Description Date Time Stamp ORGANISATION DIMENSION Organisation Dimension Identifier (PK) Corporate Office Name Region Name Branch Name Employees Name Date Time Stamp SALES FACT Time Dimension Identifier (FK) Location Dimension Identifier (FK) Product Dimension Identifier (FK) Organisation Dimension Identifier (FK) LOCATION DIMENSION Location Dimension Identifier (PK) Country Name State Code (FK) City Name Date Time Stamp Sales Dollar Date Time Stamp TIME DIMENSION Time Dimension Identifier (PK) Year Number Day of Year Quarter Number Month Number Month Name Month Day Number Week Number Day of Week Calender Date Date Time Stamp STATE LOOKUP State Code (PK) State Name Date Time Stamp MONTH LOOKUP Month Number (PK) Month Name Date Time Stamp Fig.7 Example of snowflake schema . The main disadvantage of the snowflake schema is the additional maintenance efforts needed due to the increase number of lookup tables. the example diagram shown below has 4 dimension tables. branch. ORGANIZATION. Example—Snowflake Schema In Snowflake schema. 4 lookup tables and 1 fact table.

In few organizations. Who are data warehouse users? Explain. ² In a star schema. 3. Since dimension tables hold less space. ² Whereas hierarchies are broken into separate tables in snowflake schema. Write an essay on schemas. a dimension table will have one or more parent tables. 7. These hierarchies help to drill down the data from topmost hierarchies to the lowermost hierarchies. they try to normalize the dimension tables to save space. Discuss about use of data warehouse. this Snowflake schema approach increases the number of joins and poor performance in retrieval of data. In OLAP. 4. EXERCISE 1. a dimension table will not have any parent table. 5. Explain data warehouse architecture.Data Warehouse Design | 85 respectively and shown separately. 2. 6.1.7. ² Hierarchies for the dimensions are stored in the dimensional table itself in star schema. Snowflake schema approach may be avoided. Define fact and dimension tables. How will you design OLAP cubes? . ² Whereas in a snowflake schema.3 Important Aspects of Star Schema & Snowflake Schema ² In a star schema every dimension will have a primary key.

performance. A variety of RAID controllers and disk configurations offer tradeoffs among cost. it is I/O bottleneck. ranging from 0 to 7. These topics are discussed: ² Hardware partitioning ² Software partitioning 8.0 and Windows 2000 and applications (like SQL Server) into slices (usually 16–128 KB) that are then spread across all disks participating in the RAID array.1 INTRODUCTION Data warehouses often contain large tables and require techniques both for managing these large tables and for providing good query performance across these large tables. Database architects frequently use RAID (Redundant Arrays of Inexpensive Disks) systems to overcome I/O bottlenecks and to provide higher availability. This increases disk I/O performance because the hard disks participating in the RAID array. instead of some disks becoming a bottleneck due to uneven distribution of the I/O requests. The objective of partitioning is manageability and performance related requirements. . and fault tolerance. RAID is a storage technology often used for databases larger than a few gigabytes. When a system is constrained by having limited CPU resources. RAID can be implemented in several levels. RAID can provide both performance and fault tolerance benefits. Fault Tolerance RAID also provides protection from hard disk failure and accompanying data loss by using two methods: mirroring and parity. Splitting data across physical drives like this has the effect of distributing the read/write I/O workload evenly across all physical hard drives participating in the RAID array.CHAPTER 8 PARTITIONING IN DATA WAREHOUSE 8.2 HARDWARE PARTITIONING When a system is constrained by I/O capabilities. Performance Hardware RAID controllers divide read/writes of all data from Windows NT 4. as a whole are kept equally busy. it is CPU bottleneck.

however. Mirroring can enable the system to survive at least one failed drive and may be able to support the system through failure of up to half of the drives in the mirrorset without forcing the system administrator to shut down the server and recover from the file backup. Another advantage is that mirroring provides more fault tolerance than parity RAID implementations. Both RAID 1 and its hybrid. Be ready to add disks and redistribute data across RAID arrays and/or small computer system interface (SCSI) channels as necessary to balance disk I/O and maximize performance. are usually one failed drive before the array must be taken offline and recovery from backup media must be performed to restore data. RAID 5 requires four disk I/O operations for each write. RAID 5 and its hybrids are implemented through parity. Parity information is evenly distributed among all drives participating in the RAID 5 array. . once to each side of the mirrorset.0 and Windows 2000 and RDBMS are online.Partitioning in Data Warehouse | 87 Mirroring It is implemented by writing information onto a second (mirrored) set of drives. Read I/O operation costs are the same for mirroring and parity. Read operations. Advantage The main plus of parity is cost. is often one of the most expensive components needed. General Rule of Thumb: Be sure to stripe across as many disks as necessary to achieve solid disk I/O performance. If there is a drive loss with mirroring in place. Such RAID systems are commonly referred to as “Hot Plug” capable drives. If a drive should fail. RAID 0+1 (sometimes referred to as RAID 10 or 0/1) are implemented through mirroring. which. Bear in mind that each RDBMS write to the mirrorset results in two disk I/O operations. only one additional drive is required. for a data warehouse. the data for the lost drive can be rebuilt by replacing the failed drive and rebuilding the mirror set. Advantage It offers the best performance among RAID options if fault tolerance is required. System monitor will indicate if there is a disk I/O bottleneck on a particular RAID array. a new drive is inserted into the RAID array and the data on that failed drive is recovered by taking the recovery information (parity) written on the other drives and using this information to regenerate the data from the failed drive. compared to two disk I/O operations for mirroring. Due to the additional costs associated with calculating and writing parity. Disadvantage The disk cost of mirroring is one extra drive for each drive worth of data. This essentially doubles your storage cost. Disadvantages The minus of parity are performance and fault tolerance. Parity It is implemented by calculating recovery information about data written to disk and writing this parity information on the other drives that form the RAID array. To protect any number of drives with RAID 5. Most RAID controllers provide the ability to do this failed drive replacement and re-mirroring while Windows NT 4.

These RAID controllers usually protect their caching mechanism with some form of backup power.3 RAID LEVELS Level 0 This level is also known as disk striping because of its use of a disk file system called a stripe set. Although a UPS for the server does not directly affect performance. RAID 0 improves read/write performance by spreading operations across multiple disks. The principle of these controller-based caching mechanisms is to gather smaller and potentially nonsequential I/O requests coming in from the host server and try to batch them together with other I/O requests for a few milliseconds so that the batched I/Os can form larger (32–128 KB) and maybe sequential I/O requests to send to the hard drives. RAID 0 is similar to RAID 5. This available caching with SQL Server can significantly enhance the effective I/O handling capacity of the disk subsystem. 8. In keeping with the principle that sequential and larger I/O is good for performance.88 | Data Mining and Warehousing Effect of On-Board Cache of Hardware RAID Controllers Many hardware RAID controllers have some form of read and/or write caching. 8. Rather. If the database server is also supported by an uninterruptible power supply (UPS). Disk 1 Disk 2 Disk 3 Disk 4 A0 B0 C0 D0 A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3 Fig. the RAID controller has more time and opportunity to flush data to disk in the event of power disruption. The following illustration shows RAID 0. It is not that the RAID controller caching magically allows the hard disks to process more I/Os per second. the RAID controller cache is using some organization to arrange incoming I/O requests to make best possible use of the underlying hard disks’ fixed amount of I/O processing ability. this helps produce more disk I/O throughput given the fixed number of I/Os that hard disks are able to provide to the RAID controller. This backup power can help preserve the data written in cache for some period of time (perhaps days) in case of a power outage. so that operations can be performed independently and simultaneously. Data is divided into blocks and spread in a fixed order among all disks in an array. except RAID 5 also provides fault tolerance.1 RAID level 0 . it does provide protection for the performance improvement supplied by RAID controller caching.

It also employs a disk-striping strategy that breaks a file into bytes and spreads it across multiple disks.Partitioning in Data Warehouse | 89 Level 1 This level is also known as disk mirroring because it uses a disk file system called a mirror set. RAID 1 provides fault tolerance and generally improves read performance (but may degrade write performance). Disk 1 Disk 2 Disk 3 Disk 4 Disk 5 A0 A1 A2 A3 4 parity B0 B1 B2 3 parity B4 C0 C1 2 parity C3 C4 Fig. 8. Disk mirroring provides a redundant. This strategy offers only a marginal improvement in disk utilization and read/write performance over mirroring (RAID 1).2 RAID level 1 Level 2 This level adds redundancy by using an error correction method that spreads parity across all disks. Disk 1 Disk 2 A B C D A B C D Mirror Fig. The following illustration shows RAID 1. All data written to the primary disk is written to the mirror disk. RAID 2 is not as efficient as other RAID levels and is not generally used. 8. identical copy of a selected disk.3 RAID level 2 D0 1 party D2 D3 D4 0 party E1 E2 E3 E4 .

However. Like RAID 3. the error correction method requires only one disk for parity data. RAID 4 is not as efficient as other RAID levels and is not generally used.4 RAID level 5 Level 10 (1 + 0) This level is also known as mirroring with striping. This level uses a striped array of disks. RAID 3 also is rarely used. which are then mirrored to another identical set of striped disks. RAID 10 provides the performance benefits of disk striping with the disk redundancy of mirroring. Disk 1 Disk 2 Disk 3 Disk 4 Disk 1 Disk 2 Disk 3 Disk 4 A0 B0 C0 D0 A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3 A0 B0 C0 D0 A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3 Fig. The following illustration shows RAID 10. It keeps user data separate from error-correction data. The striped array of disks is then mirrored using another set of four striped disks. a striped array can be created using four disks. It is similar to RAID 4 because it stripes the data in large blocks across the disks in an array. Level 5 Also known as striping with parity. Striping with parity offers better performance than disk mirroring (RAID 1). this level is the most popular strategy for new designs. The following illustration shows RAID 5. . read performance degrades (for example. Use of disk space varies with the number of data disks.90 | Data Mining and Warehousing Level 3 This level uses the same striping method as RAID 2. when a stripe member is missing. RAID 10 provides the highest read/write performance of any of the RAID levels at the expense of using twice as many disks. RAID 3 provides some read/write performance improvement. Data redundancy is provided by the parity information. It differs in how it writes the parity across all the disks. The data and parity information are arranged on the disk array so the two are always on different disks. but the error correction method requires only one disk for parity data. For example. Level 4 This level employs striped data in much larger blocks or segments than RAID 2 or RAID 3. when a disk fails). 8. RAID 5 is one of the most commonly used RAID configurations.

but cost more in terms of disks required.5 Comparison of different RAID levels . 8. RAID 5 costs less than RAID 1 or RAID 0+1 but provides less fault tolerance and less write performance. RAID 0 is typically used only for benchmarking or testing. it should never be used in a production environment. The write performance of RAID 5 is only about half that of RAID 1 or RAID 0+1 because of the additional I/O needed to read and write parity information.Partitioning in Data Warehouse | 91 As mentioned above. RAID 0 RAID 1 RAID 5 RAID 0 + 1 Data Fault tolerance information Fig. When cost of hard disks is not a limiting factor. RAID 1 and RAID 0+1 offer the best data protection and best performance among RAID levels. and it is not recommended for development environments. Because RAID 0 provides no fault tolerance protection. The best disk I/O performance is achieved with RAID 0 (disk striping with no fault tolerance protection). RAID 1 or RAID 0+1 are the best choices in terms of both performance and fault tolerance.

92 | Data Mining and Warehousing

Many RAID array controllers provide the option of RAID 0+1 (also referred to as RAID 1/ 0 and RAID 10) over physical hard drives. RAID 0+1 is a hybrid RAID solution. On the lower level, it mirrors all data just like normal RAID 1. On the upper level, the controller stripes data across all of the drives (like RAID 0). Thus, RAID 0+1 provides maximum protection (mirroring) with high performance (striping). These striping and mirroring operations are transparent to Windows and RDBMS because they are managed by the RAID controller. The difference between RAID 1 and RAID 0+1 is on the hardware controller level. RAID 1 and RAID 0+1 require the same number of drives for a given amount of storage. For more information on RAID 0+1 implementation of specific RAID controllers, contact the hardware vendor that produced the controller. The Fig. 8.5 shows differences between RAID 0, RAID 1, RAID 5, and RAID 0+1. Note: In the illustration above, in order to hold four disks worth of data, RAID 1 (and RAID 0+1) need eight disks, whereas Raid 5 only requires five disks. Be sure to involve your storage vendor to learn more about their specific RAID implementation.

² ² ² ² ² Range Partitioning Hash Partitioning Index Partitioning Composite Partitioning List Partitioning

Range Partitioning
Range partitioning maps data to partitions based on ranges of partition key values that you establish for each partition. It is the most common type of partitioning and is often used with dates. For example, you might want to partition sales data into weekly, monthly or yearly partitions. Range partitioning maps rows to partitions based on ranges of column values. Range partitioning is defined by the partitioning specification for a table or index. PARTITION BY RANGE (column_list) and by the partitioning specifications for each individual partition: VALUES LESS THAN (value_list) where: column_list is an ordered list of columns that determines the partition to which a row or an index entry belongs. These columns are called the partitioning columns. The values in the partitioning columns of a particular row constitute that row’s partitioning key. value_ list is an ordered list of values for the columns in the column list. Only the VALUES LESS THAN clause is allowed. This clause specifies a non-inclusive upper bound for the partitions. All partitions, except the first, have an implicit low value specified by the VALUES LESS THAN literal on the previous partition. Any binary values of the partition key equal to or higher than this literal are added to the next higher partition. Highest partition being where MAXVALUE literal is defined. Keyword, MAXVALUE, represents a virtual infinite value that sorts higher than any other value for the data type, including the null value. The statement below creates a table sales_range_part that is range partitioned on the sales_date field using Oracle.

Partitioning in Data Warehouse | 93

CREATE TABLE sales_range_part (sales_man_id NUMBER(6), emp_name VARCHAR2(25), sales_amount NUMBER(10), sales_date DATE) PARTITION BY RANGE(sales_date) ( PARTITION sales_jan2006 VALUES LESS THAN(TO_DATE(’02/01/2006',’MM/DD/YYYY’)), PARTITION sales_feb2006 VALUES LESS THAN(TO_DATE(’03/01/2006',’MM/DD/YYYY’)), PARTITION sales_mar2006 VALUES LESS THAN(TO_DATE(’04/01/2006',’MM/DD/YYYY’)), PARTITION sales_apr2006 VALUES LESS THAN(TO_DATE(’05/01/2006',’MM/DD/YYYY’)),);

Hash Partitioning
Hash partitioning maps data to partitions based on a hashing algorithm that Oracle applies to a partitioning key that you identify. The hashing algorithm evenly distributes rows among partitions, giving partitions approximately the same size. Hash partitioning is the ideal method for distributing data evenly across devices. Hash partitioning is a good and easy-to-use alternative to range partitioning when data is not historical and there is no obvious column or column list where logical range partition pruning can be advantageous. Oracle uses a linear hashing algorithm and to prevent data from clustering within specific partitions, you should define the number of partitions by a power of two (for example, 2, 4, 8). The statement below creates a table sales_hash, which is hash partitioned on the salesman_id field. data1, data2, data3, and data4 are tablespace names. CREATE TABLE sales_hash (salesman_id NUMBER(5), emp_name VARCHAR2(25), sales_amount NUMBER(10), week_no NUMBER(2)) PARTITION BY HASH(salesman_id) PARTITIONS 4 STORE IN (data1, data2, data3, data4);

List Partitioning
List partitioning enables you to explicitly control how rows map to partitions. You do this by specifying a list of discrete values for the partitioning column in the description for each partition. This is different from range partitioning, where a range of values is associated with a partition and with hash partitioning, where you have no control of the row-to-partition mapping. The advantage of list partitioning is that you can group and organize unordered and unrelated sets of data in a natural way. CREATE TABLE sales_list (salesman_id NUMBER(5), emp_name VARCHAR2(25), sales_area VARCHAR2(20), sales_amount NUMBER(10), sales_date DATE) PARTITION BY LIST(sales_area) ( PARTITION sales_south VALUES IN(‘Delhi’, ‘Mumbai’,’Kolkatta’), PARTITION sales_north VALUES IN (‘Bangalore’, ‘Chennai’, ‘Hyderabad’));

94 | Data Mining and Warehousing

Composite Partitioning
Composite partitioning combines range and hash partitioning. Oracle first distributes data into partitions according to boundaries established by the partition ranges. Then Oracle uses a hashing algorithm to further divide the data into subpartitions within each range partition. CREATE TABLE sales( s_productid NUMBER, s_saledate DATE, s_custid NUMBER, s_totalprice NUMBER) PARTITION BY RANGE (s_saledate) SUBPARTITION BY HASH (s_productid) SUBPARTITIONS 16 (PARTITION sal98q1 VALUES LESS THAN (TO_DATE(’01-APR-1998', ‘DD-MON-YYYY’)), PARTITION sal98q2 VALUES LESS THAN (TO_DATE(’01-JUL-1998', ‘DD-MON-YYYY’)), PARTITION sal98q3 VALUES LESS THAN (TO_DATE(’01-OCT-1998', ‘DD-MON-YYYY’)), PARTITION sal98q4 VALUES LESS THAN (TO_DATE(’01-JAN-1999', ‘DD-MON-YYYY’)))

Index Partitioning
You can choose whether or not to inherit the partitioning strategy of the underlying tables. You can create both local and global indexes on a table partitioned by range, hash, or composite methods. Local indexes inherit the partitioning attributes of their related tables. For example, if you create a local index on a composite table, Oracle automatically partitions the local index using the composite method.

1. Why we need partition? Explain. 2. Explain different types of partition with suitable examples. 3. Write a short note on RAID technology.

A data warehouse provides an opportunity for slicing and dicing that cube along each of its dimensions. In addition to a relational/multidimensional database. There are three types of data warehouses: 1. 3. ODS (Operational Data Store) . an OLAP engine. data is refreshed in near real time and used for routine business activity. client analysis tools. Data Mart . A data mart cannot stand alone. Data warehouses and data marts are built on dimensional data modeling where fact tables are connected with dimension tables. It contains summarized data that the user group can easily understand.Datamart is a subset of data warehouse and it supports a particular region. and other applications that manage the process of gathering data and delivering it to business users.An enterprise data warehouse provides a central database for decision support throughout the enterprise. Because each data warehousing effort is unique. it requires a data warehouse.1 INTRODUCTION A data warehouse is a relational/multidimensional database that is designed for query and analysis rather than transaction processing.2 DATA MART A data warehouse is a cohesive data model that defines the central data repository for an organization. This is most useful for users to access data since a database can be visualized as a cube of several dimensions. a data warehouse environment often consists of an ETL solution. process. Each data mart is a collection of tables organized according to the particular requirements of a user or group of users. Retrieving a collection of different kinds of data from a “normalized” .This has a broad enterprise wide scope. your company’s data warehousing environment may differ slightly from what we are about to introduce. Enterprise Data Warehouse . and apply. It separates analysis workload from transaction workload and enables a business to consolidate data from several sources. 2. business unit or business function. A data mart is a data repository for a specific user group. 9. A data warehouse usually contains historical data that is derived from transaction data. but unlike the real entertprise data warehouse.CHAPTER 9 DATA MART AND META DATA 9.

And however the marts are organized initially. has millions of customers and the lines of business of the enterprise are savings. variable measurements. and physical attributes of data. .3 META DATA For Example: In order to store data. it may not even make sense. This organization does not have to follow any particular inherent rules or structures. ² Tools to convert models into data marts quickly. The following example explains how the data is integrated from source systems to target systems. and video tapes all next to each other. Data Warehouse Data Mart1 Data Mart2 Data Mart N Fig. This means that the creation of data marts requires: ² Understanding of the business involved. over the years. So source systems will be different in naming conventions. the requirements are almost certain to change once the user has seen the implications of the request. ² Sufficient facility with database modeling and design to produce new tables quickly. encoding structures. The notion of a “mart” suggests that it is organized for the ultimate consumers — with the potato chips. Indeed.96 | Data Mining and Warehousing warehouse can be complex and time-consuming. many application designers in each branch have made their individual decisions as to how an application and database should be built. 9. Hence the need to rearrange the data so they can be retrieved more easily.1 Data warehouse and data marts 9. and loans. ² Responsiveness to the user’s stated objectives. Consider a bank that has got several branches in several countries.

the job of the analyst is much more difficult and the work required for analysis is compounded significantly. column names. This is how data from various source systems is integrated and accurately stored into the data warehouse. how it relates to other information.. attribute name. Example of Target Data (Data Warehouse) Target system Record #1 Record #2 Record #3 Attribute name Customer Application Date Customer Application Date Customer Application Date Column name CUSTOMER_ APPLICATION_DATE CUSTOMER_ APPLICATION_DATE CUSTOMER_ APPLICATION_DATE Datatype DATE DATE DATE Values 01112005 01112005 01112005 In the above example of target data. We intend to create a design process for creating a data warehouse through the development of a metadata system. the user of the data warehouse must have access to meta data that is accurate and up to date. 0) 11012005 DATE DATE 11012005 01NOV2005 In the aforementioned example. datatype and values are entirely different from one source system to another. The meta data system we intend to create will provide a framework to organize and design the data warehouse. In order to be effective. Using a systems engineering approach.Data Mart and Meta Data | 97 Example of Source Data System name Source System 1 Source System 2 Source System 3 Attribute name Customer Application Date Customer Application Date Application Date Column name CUSTOMER_ APPLICATION_DATE CUST_APPLICATION_ DATE APPLICATION_DATE Datatype Values NUMERIC (8. where it comes from. A data warehouse stores current and historical data from disparate operational systems (i. This inconsistency in data can be avoided by integrating the data into a data warehouse with good standards. attribute names.e. transactional databases) into one consolidated system where data is cleansed and restructured to support data analysis. The topic of standardizing meta data across various products and applying a systems engineering approach to this process in order to facilitate data warehouse design is what this project intend to address. this process will be used initially to help define the data warehouse requirements and it will then be used iteratively during the life of the data warehouse to update and integrate new dimensions of the warehouse. column name. where it stored. Meta data is literally “data about data”. Without a good source of meta data to operate from. and datatypes are consistent throughout the target system. . and how it is related to the business. It describes the kind of information in the warehouse.

The data warehouse presents an alternative in data processing that represents the integrated information requirements of the enterprise. Destination . Effective tools must be used to properly store and use the meta data generated by the various systems. Name is its local field name. 9. Location is both the system comes from and the object contains it.refers to the breakup of data into separate physical units that can be handled independently. For each source data field the following information is required: Source field Unique identifier. Destination Unique identifier. type.refers to the level of detail held in the unit of data ² Partitioning of data . The destination field needs to be described in a similar way to the source. Base data is used to distinguish between the original data load destination and any copies. It will also depend on the number of source systems and the type of each system. when moving it from source to data warehouse. Some specific challenges which involve the physical design tradeoffs of a data warehouse include: ² Granularity of data . Whether or not you need to store any meta data for this process will depend on the complexity of the transformations that are to be applied to the data. name. Name is the designation that the field will have in the base data within the data warehouse. type. A unique identifier is required to avoid any confusion occurring between two fields of the same name from different sources. table name. Type is the storage type of data. location (system. type. Meta data itself is arguably the most critical element in effective data management. Name. data has aggregated in the corporate world as a by-product of transaction processing and other uses of the computer.1 Data Transformation and Load Meta data may be used during transformation and load to describe the source data and any changes that need to be made. A unique identifier is needed to help avoid any confusion between commonly used field names.98 | Data Mining and Warehousing Understanding the role of the enterprise data model relative to the data warehouse is critical to a successful design and implementation of a data warehouse and its meta data management. ² Performance issues ² Data structures inside the warehouse ² Migration For years. For the most part. the data that has accumulated has served only the immediate set of requirements for which is was initially intended. location describe where the data comes from. name. It is designed to support analytical processing for the entire business organization. object).3. These trends and patterns are absolutely vital to the vision of management in directing the organization. Analytical processing looks across either various pieces of or the entire enterprise and identifies trends and patterns otherwise not apparent.

table (columns) To this end we will need to store meta data like the following for each field.3. Transformation needs to be applied to turn the source data into the destination data. 9. varchar. average. The other attributes are module name and syntax. type) Indexes Columns (name. reference identifier. field name. syntax). Name is a unique identifier that differentiates this from any other similar transformations. type) Views Columns (name.Data Mart and Meta Data | 99 type is the database data type. Aggregation Aggregation name. number. type. language (module name. type) Constraints Name. Language attribute contains the name of the language that the transformation written in. columns (column_name. Field Unique identifier. aggregation). description. Table Table name. Table name is the name of the table field will be part of. and they require the following meta data to be stored.2 Data Management Meta data is required to describe the data as it resides in the data warehouse. Meta data is needed for all of the following: Table Columns (name. sum . max. For each table the following information needs to be stored. The functions listed below are examples of aggregation functions: Min. columns (column_name. such as clear. Transformation(s) Name. reference identifier) Aggregations are a special case of tables. Every object in the database needs to be described. This is needed by the warehouse manager to allow it to track and control all data movements.

. table name (range allowed. technology.100 | Data Mining and Warehousing Partitions are subsets of a table. 3. Partition Partition name. depending on the nature. Moreover. What is data mart? Give example. EXERCISE 1. partition key(reference identifier)) 9. Inc. Combining this with the mapping of each column in a data mart to an attribute of the data model provides direct mapping of original data to the view of them seen by users. and sites of those systems.4 LEGACY SYSTEMS The task of documenting legacy systems and mapping their columns to those of the central data warehouse is the worst part of any data warehouse project. Essential Strategies. 2. range contained. has extended Oracle’s Designer/2000 product to allow for reverse engineering of legacy systems—specifically providing the facility for mapping each column of a legacy system to an attribute of the corporate data model. What is legacy system? Explain data management in meta data. Define meta data. but they also need the information on the partition key and the data range that resides in the partition. the particular characteristics of the project vary. 4.

Cold Backup—Backup that is taken while the database is completely shut down. then treat it accordingly. Human errors. Consider how these important issues affect backup plan: ² Realize that the restore is almost more important than the backup ² Create a hot backup ² Move your backup offline ² Perfect the connection between your production and backup servers ² Consider a warm backup ² Always reassess the situation .If such failure affects the operation of a database system . Software. 10. system failure and the other unfrozen disasters can make your work vanish into digital nothingness in no time. In a multiinstance environment. The terminology comes from the fact that the database engine is up and running hence hot. It make sense backup your data on a regular basis . is required to recover the data base and return to normal operation as quick as possible. There are special requirements that need to be observed if the hot backup is to be valid. This includes all database data files and the journal files. Hot Backup—Any backup that is not cold is considered to be hot.1 INTRODUCTION One of the most important part is backup and recovery in a data warehouse.CHAPTER 10 BACKUP AND RECOVERY OF THE DATA WAREHOUSE 10. Hardware.2 TYPES OF BACKUP The definitions below should help to clarify the issue. If your data is precious to you. These definitions apply to the back up of the database. Complete Backup—The entire database is backed up at the same time. all instances of that database must be shut down. There are utilities that can backup entire partitions along with the operating system and everything else. Partial Backup—Any backup that is not complete. Process. These will vary from RDBMS to RDBMS.

This provides for efficient operations that can scale up to handle very large volumes of data. if lost. Furthermore. ² Oracle includes support for incremental backup and recovery. For other backup products designed for local backup. you should at least do periodic test restores. and recovery operations are maintained by the server in a recovery catalog and automatically used as part of these operations. Cold backups shut down the database. Even though most of the data comes from operational systems originally. enabling operations to be completed efficiently within times proportional to the amount of changes. and provides tight interfaces to industry-leading media management subsystems. This reduces administrative burden and minimizes the possibility of human errors. can be backed up and restored independently of the other partitions of a table. The major types of instance recovery are cold restore. Backup and recovery problems are most likely to occur at this level. Some of the highlights are: ² Details related to backup. This will also provide you with a “hot” backup that is ready to be used. As operational data ages. when placed in their own tablespaces.102 | Data Mining and Warehousing 10. should your production database ever crash. Individual partitions. Keeping a local backup allows you to save data way in excess of the storage that you purchase for online backup.1 Create a Hot Backup The best solution for critical systems is always to restore the backup once it has been made onto another machine. If you omit any of these files. restore. you have not made a successful backup of the database. ² Backup and recovery operations are fully integrated with partitioning. We allow you to back up any number of computers to your account. You only pay a small one time fee for each additional computer. time-based recovery. Hot backups take backups while the database is functioning. data warehouses often contain external data that. Open platforms for more hardware options & enterprise-level platforms. Each type of backup has its advantages and disadvantages. There are also supplemental backup methods.3 BACKUP THE DATA WAREHOUSE Oracle RDBMS is basically a collection of physical database files. but it may still exist in the data warehouse. you cannot always rebuild data warehouses in the event of a media failure (or a disaster).3. Putting in place a backup and recovery plan for data warehouses is imperative. All computers will share the common pool of storage space. 10. . and online redo log files. full database recovery. such as exports. may have to be purchased. and cancel-based recovery. ² The backup and recovery technology is highly scalable. Three types of files must be backed up: database files. Oracle provides a server-managed infrastructure for backup. control files. If you can’t afford this higher cost. it is removed from the operational databases. restore. safer operations at terabyte scale. and recovery tasks that enables simpler. rather than the overall size of the database.

you might set the hot backup within a minute of the primary server. Ideally you should keep a copy of this data at some location other than your home or place of business.3. to enable users to make backup copies of their data. too. except there’s one thing you still need to worry about. This is easily done with an online backup service. a leader in the backup business. If you use your computer for business or personal use. another hard drive. 10. they too will be lost..2 Consider a Warm (Not Hot) Backup After you install hot backups and build a solid network. For example. You also need to have multiple versions of your information backed up. Simple data backup can be done by saving your data to multiple locations. you propagate that change to your backup server. For example. fairly recent but not as recent as your hot backup. That way. To make this task easier. You need to make backups of your important data to prevent a total loss in the event of any kind of system failure or human error. But that very same close binding also comes with its own dangers. but deliberately keep your warm backup 30 minutes behind your primary. If every second of downtime is expensive. If your house burns down and your backups are sitting on the shelf next to your computer. say an administrator accidentally drops the accounts table in the database. this actually happened at a client of mine. can you relax? Sort of. (I’m not making this up. Hot backups are great because they are extremely current.3 Tape Backup Storage ² ² ² ² ² Tape backup of customers’ data on SAN Tape backup of Veritas disk storage Tape backup of customers’ data direct from network Tape archiving Off-site tape storage 10. you need to always have a backup of your data files in the event of software. The minute data is changed on the primary server.Backup and Recovery of the Data Warehouse | 103 10.) The inadvertent deletion will immediately get propagated to the backup.3. If you are working on a spreadsheet over a period of several days. SureWest Broadband has licensed Backup Software from NovaStor. copying the data from its original location to removable media.3. it’s best to keep a snapshot of every day’s work. you only have to restore the last several transaction logs rather than start from scratch. or another computer’s hard drive on the network. hardware or human error.e.4 What is SureWest Broadband’s Online Backup Service? It is Internet-based services that allows computer users to routinely backup and recover their data using a secure connection and safely store it in the SureWest Broadband Data Storage Warehouse until it is needed. Consider what happens if someone accidentally deletes the wrong data. Most users . you should consider creating a copy of the database that is kept “warm”—i.

but it does so with significant exposure to data loss in the event of a system failure. the loss of committed transactions) . Online backup software is designed to routinely copy your important files to a private. create index or bulk copy). When using the simple recovery model. 10. and simple. Knowledgeable administrators can use this recovery model feature to significantly speed up data loads and bulk operations. The model for a database can be changed after the database has been created. the transactions are truncated from the log upon checkpoint. they are saved into a new file and compressed into what is known as a patch. so a new type of backup system was needed. the amount of exposure to data loss varies with the model chosen. data is recoverable only to the last (most recent) full database or differential backup. If you have a working Internet connection on your computer. When the differences are extracted from the two files. It’s known as online backup. It does this at the expense of some flexibility of point-in-time recovery.104 | Data Mining and Warehousing do not want to be bothered with managing the tasks necessary for maintaining their own backups. in this model. This creates the potential for data loss. The patch file is often 85% to 1010. you are deciding among the following business requirements: ² Performance of large-scale operations (for example. Each recovery model addresses a different need. you can use SureWest Broadband’s Online Backup Service to keep your important files safe from disaster on a daily basis. it is truncated and reused. It is imperative that the risks be thoroughly understood before choosing a recovery model. However. ² Simple recovery provides the highest performance and lowest log space consumption.4 DATA WAREHOUSE RECOVERY MODELS The model chosen can have a dramatic impact on performance. secure location on the Internet by transmitting your data over your existing Internet connection. There are three recovery models: full. After the log space is no longer needed for recovery from server failure (active transactions).10% smaller than the file which the patch was extracted from originally. bulk-logged. ² Full recovery provides the most flexibility for recovering databases to an earlier point in time. and protection against data loss. ² Bulk-logged recovery provides higher performance and lower log space consumption for certain large-scale operations (for example. space utilization (disk or tape). especially during data loads. Trade-offs are made depending on the model you chose. The trade-offs that occur pertain to performance.3. 10. Transaction log backups are not usable for recovering transactions because. All of your data is encrypted before it is sent to our storage network to protect your privacy.5 What is a FastBIT Backup? The FastBIT patching process is the core technology behind NovaStor’s speedy backup program. The patching process involves the comparison of two different versions of the same file and extracting the differences between the files. index creation or bulk loads) ² Data loss exposure (for example. When you choose a recovery model. The recovery model of a new database is inherited from the model database when the new database is created.

one model may be more appropriate than another. Can recover to any point in time. The following table provides helpful information. Explain data warehouse recovery model. Recover to point in time? Can recover to the end of any backup. or bulk operations occurred since the most recent log backup. changes since the most recent log backup must be redone. Work loss exposure Changes since the most recent database or differential backup must be redone Normally none. no work is lost. How will you backup the data warehouse? Explain. Can recover to an arbitrary point in time (for example.Backup and Recovery of the Data Warehouse | 105 ² Transaction log space consumption ² Simplicity of backup and recovery procedures Depending on what operations you are performing. . Permits high-performance bulk copy operations. If the log is damaged. Then changes must be redone. Minimal log space is used by bulk operations. What are the types of backup? 3. prior to application or user error). Before choosing a recovery model. Can recover to the end of any backup. consider the impact it will have. 2. Then changes must be redone. Otherwise. Recovery model Simple Benefits Permits high-performance bulk copy operations. Full Bulk-Logged If the log is damaged. changes since that last backup must be redone. No work is lost due to a lost or damaged data file. Reclaims log space to keep space requirements small. EXERCISE 1.

1 illustrates the tuning method: . resolve tuning issues in the order listed: from the design and development phases through instance tuning. These steps are prioritized in order of diminishing returns: steps with the greatest effect on performance appear first. Furthermore. Different tuning strategies vary in their effectiveness.1 INTRODUCTION A well planned methodology is the key to success in performance tuning. Fig. Step Step Step Step Step Step Step Step Step Step 1: Tune the Business Rules 2: Tune the Data Design 3: Tune the Application Design 4: Tune the Logical Structure of the Database 5: Tune Database Operations 6: Tune the Access Paths 7: Tune Memory Allocation 8: Tune I/O and Physical Structure 9: Tune Resource Contention 10: Tune the Underlying Platform(s) After completing these steps. such as online transaction processing systems and decision support systems. so additional passes through the tuning process may be useful.2 PRIORITIZED TUNING STEPS The following steps provide a recommended method for tuning an Oracle database. Performance gains made in later steps may pave the way for further improvements in earlier steps. likely require different tuning methods. reassess your database performance and decide whether further tuning is necessary.CHAPTER 11 PERFORMANCE TUNING AND FUTURE OF DATA WAREHOUSE 11. 11. Tuning is an iterative process. systems with different purposes. 11. therefore. For optimal results.

11. The tuning method .1.Performance Tuning and Future of Data Warehouse | 107 1 Tune the business rules 2 Tune the data design 3 Tune the application design 4 Tune the logical structure of the database 5 Tune database operations 6 Tune the access paths 7 Tune memory allocation 8 Tune the I/O and physical structure 9 Tune the resource contention 10 Tune the underlying platform(s) Fig.

3. 5. 4. you may need to return from any step to any previous step. in step 5 you may rewrite some of your SQL statements. 11.108 | Data Mining and Warehousing Decisions you make in one step may influence subsequent steps. These SQL statements may have significant bearing on parsing and caching issues addressed in step 7. For example. which is tuned in step 8. 7.3 CHALLENGES OF THE DATA WAREHOUSE 1.4 BENEFITS OF DATA WAREHOUSING 1. 3. Also. which is tuned in step 7. software staffing 32% Selling to management 26% Training users 16% Managing expectations 11% Managing change 8% 11. 2. Better data quality 63% Better competitive data 61% Direct end-user access 32% Cost reduction/higher productivity 32% More timely data access 24% Better availability of systems 16% Supports organization change 8% 11. 4. 7. Technical challenges 42% Data management 40% Hardware. 6. 6. Although the figure shows a loop back to step 1. disk I/O. 5. depends on the size of the buffer cache. 2.5 FUTURE OF THE DATA WAREHOUSE ² Peta byte system (1 PB = 1024 TB) ² Size of the database grows to a Very Large Database (VLDB) to ELDB (Extremely Very Large Data Base) ² Integration and manipulation of non-textual (multimedia) and textual data ² Building and running ever-larger data warehouse system ² Web enabled application grows ² Handle vast quantities of multiformat data ² Distributed databases will be used ² Cross database integrity ² Use of middleware and multiple tiers .

3. Data Mart 2… Data Mart n Business Intelligence (B. Transformation and Load Process Data Warehouse Data Mart1. Explain tuning the data warehouse. Business and Internal] Fig.I) l Reports [Organization .6 NEW ARCHITECTURE OF DATA WAREHOUSE Business Modeling l Business Process l Data Models Data Modeling Source data Model l Data Warehouse Models l Source Design * RDBMS * Source Systems * Files * ERP [SAP] ETL Procedure Designing Extraction. Draw the architecture diagram of data warehouse.2 Data warehouse architecture diagram EXERCISE 1.Performance Tuning and Future of Data Warehouse | 109 11. 2. What is the future scope of data warehouse? . 11.

Aggregation: The process of consolidating data values into a single value. Attributes represent logical groupings that enable end users to select data based on like characteristics. Byte: A standard unit of storage. Basically. . The general form of an association rule is X1. quarter and yearly sales.Xn predict Y. This means that the attributes X1. Association Rules: rules that state a statistical correlation between the occurrence of certain attributes in a database table. In RDBMS. typically as part of the ETL process.Appendix A GLOSSARY Aggregate: Summarized data. The data can then be referred to as aggregate data. the week data could be aggregated to the month level. Classification: A subdivision of a set of examples into a number of classes. Aggregation is synonymous with summarization. Cold Backup: A database backup taken while the database is shut down. X2. See also ETL.…. Bit(s): acronym for binary digit. 0 or 1. Attribute: A descriptive characteristic of one or more levels. A byte consists of 8 bits. a byte is a sufficient to store a single character. X2. with the database is shut down. sales data could be collected on a daily basis and then be aggregated to the week level. Cleansing: The process of resolving inconsistencies and fixing the anomalies in source data. A unit of computer storage that can store 2 values. and produces some value or set of values as output. Business Intelligence Tools: Software that enables business users to see and use large amounts of complex data.…. unit sales of a particular product could be aggregated by day. month. For example.Xn =>y. For example. Coding: Operations on a database to transform or simplify data in order to prepare it for a machine-learning algorithm. Algorithm: Any well-defined computational procedure that takes some value or set of values as input. Clickstream Data: Data regarding web browsing Cluster: A tightly coupled group of SMP machines. an attribute is a column in a dimension that characterizes elements of a single level. and so on. and aggregate data is synonymous with summary data.

Glossary | 111

Common Warehouse Meta Data (CWM): A repository standard used by Oracle data warehousing, decision support, and OLAP tools including Oracle Warehouse Builder. The CWM repository schema is a standalone product that other products can share each product owns only the objects within the CWM repository that it creates. Confidence: Given the association rule X1, X2…Xn=>Y, the confidence is the percentage of records for which Y holds, within the group of records for which X1, X2,…Xn holds. Data: Any symbolic representation of facts or ideas from which information can potentially be extracted. Data Selection: The stage of selecting the right data for a KDD process. Data Source: A database, application, repository, or file that contributes data to a warehouse. Data Mart: A data warehouse that is designed for a particular line of business, such as sales, marketing, or finance. In a dependent data mart, the data can be derived from an enterprise-wide data warehouse. In an independent data mart, data can be collected directly from sources. See also data warehouse. Data Mining: The actual discovery phase of a knowledge discovery process. Data Warehouse: A relational database that is designed for query and analysis rather than transaction processing. A data warehouse usually contains historical data that is derived from transaction data, but it can include data from other sources. It separates analysis workload from transaction workload and enables a business to consolidate data from several sources. In addition to a relational database, a data warehouse environment often consists of an ETL solution, an OLAP engine, client analysis tools, and other applications that manage the process of gathering data and delivering it to business users. See also ETL, OLAP. Denormalize: The process of allowing redundancy in a table so that it can remain flat. Contrast with normalize. Derived Fact (or Measure): A fact (or measure) that is generated from existing data using a mathematical operation or a data transformation. Examples include averages, totals, percentages, and differences. Decision Trees: A decision tree consists of nodes and branches, starting from a single root node. Each node represents a test or decision. Depending on the outcome of the decision, one chooses a certain branch and when a terminal node(or leaf) is reached a decision on a class assignment is made. Deep Knowledge: Knowledge that is hidden within a database and can only be recovered if one is given certain clues. Dimension: A structure, often composed of one or more hierarchies, that categorizes data. Several distinct dimensions, combined with measures, enable end users to answer business questions. Commonly used dimensions are customer, product, and time. Dimension Data: Data stored externally to the fact data, which contains expanded information on specific dimensions of the fact data. Dimension Table: Part of star schema. A table which holds dimension data.

112 | Data Mining and Warehousing

Drill: To navigate from one item to a set of related items. Drilling typically involves navigating up and down through the levels in a hierarchy. When selecting data, you can expand or collapse a hierarchy by drilling down or up in it, respectively. See also drill down, drill up. Drill Down: To expand the view to include child values that are associated with parent values in the hierarchy. (See also drill, drill up) Drill Up: To collapse the list of descendant values that are associated with a parent value in the hierarchy. Enrichment: A stage of the KDD process in which new data is added to the existing selection. Element: An object or process. For example, a dimension is an object, a mapping is a process, and both are elements. ETL: Extraction, transformation, and loading. ETL refers to the methods involved in accessing and manipulating source data and loading it into a data warehouse. The order in which these processes are performed varies. Note that ETT (extraction, transformation, transportation) and E T M (extraction, transformation, move) are sometimes used instead of ETL. (See also data warehouse, extraction, transportation) Extraction: The process of taking data out of a source as part of an initial phase of ETL Fact Data: The fact data is the basic core data that will be stored in the data warehouse. It is called fact data because it will correspond to business-related facts such as call record data for a telco, or account transaction data for a bank. Fact Table: Part of a star schema. The central table that contains the fact data. File System: An area of space set aside to contain files. The file system will contain a root directory in which other directories or files can be created. File-to-table Mapping: Maps data from flat files to tables in the warehouse. Genetic Algorithm: A class of machine-learning algorithm that is based on the theory of evaluation. Granularity: The level of detail of the facts stored in a data warehouse. Hidden Knowledge: Knowledge that is hidden in a database and that cannot be recovered by a simple SQL query. More advanced machine-learning techniques involving many different SQL queries are necessary to find hidden knowledge. Hierarchy: A logical structure that uses ordered levels as a means of organizing data. A hierarchy can be used to define data aggregation; for example, in a Time dimension, a hierarchy might be used to aggregate data from the month level to the quarter level to the year level. A hierarchy can also be used to define a navigational drill path, regardless of whether the levels in the hierarchy represent aggregated totals. Inductive Learning: Learning by generalizing form examples. Information: Meaningful data. KDD: Knowledge Discovery in Databases: The non-trivial extraction of implicit, previously unknown and potentially useful information from data. Knowledge: A collection of interesting and useful patterns in a database.

Glossary | 113

Learning: An individual learns how to carry out a certain task by making a transition from a situation in which the task cannot be carried out to a situation in which the same task can be carried out under the same circumstances. Machine Learning: A sub-discipline of computer science that deals with the design and implementation of learning algorithms. Materialized View: A pre-computed table comprising aggregated and/or joined data from fact and possibly dimension tables. Also known as a summary or aggregate table. Meta Data: Data about data. Meta data is data that describes the source, location and meaning of another piece of data. For example, the schema design of a data warehouse is typically stored in a repository as metadata, which is used to generate scripts used to build and populate the data warehouse. A repository contains meta data. Examples include: for data, the definition of a source to target transformation that is used to generate and populate the data warehouse; for information, definitions of tables, columns and associations that are stored inside a relational modeling tool; for business rules, discount by 10 percent after selling 1,000 items. Multi-dimensional Knowledge: A table with n independent attributes can be seen as an ndimensional space. Some regularities represented in such a table cannot easily be detected when the table has a standard two-dimensional representation. WE have to explore the relationship Neural Networks: A class of learning algorithm consisting of multiple nodes that communicate through their connecting synapses. Neural networks initiate the structure of biological nervous systems. Normalize: In a relational database, the process of removing redundancy in data by separating the data into multiple tables. Contrast with denormalize. Operational Data Store (ODS): The cleaned, transformed data from a particular source database. OLAP: Online analytical Processing OLTP: Online transaction processing. An OLTP system is designed to handle a large volume of small predefined transactions. Online Backup: A database backup taken while the database is open and potentially in use. The RDBMS will need to have special facilities to ensure that data in the backup is consistent. Overnight Processing: Any processing that needs to be done to accomplish the daily loading, cleaning, aggregation, maintenance and backup of data, in order to make that new data available to the users for the following day. Patterns: Structures in a database that are statistically relevant. Prediction: The result of the application of a theory or a rule in a specific case. Query Tools: Tools designed to query a database. Shallow Knowledge: The information stored in a database that can be retrieved with a single query. Snowflake Schema: A variant of the star schema where each dimension can have its own dimensions.

114 | Data Mining and Warehousing Star Schema: A logical structure that has a fact table in the centre with dimension table radiating off of this central table. the support is the percentage of records for which X1…. Validation: The process of verifying metadata definitions and configuration parameters.Xn and Y both hold. . X2. Verification: The validation of a theory on the basis of a finite number of examples. Supervised Algorithms: Algorithms that need the control of a human operator during their execution. The fact table will generally be extremely large in comparison to its dimension tables. Transportation: The process of moving copied or transformed data from a source to a data warehouse. Support: Given the association rule X1. Unsupervised Algorithms: The opposite of supervised algorithms. There will be a foreign key in the fact table for each dimension table. Star Flake Schema: This is a hybrid structure that contains a mix of star and snowflake schemas. Versioning :The ability to create new versions of a data warehouse project for new requirements and changes. Statistical Techniques: Techniques used in statistics.…Xn => Y. Some dimensions may be represented in both forms to cater different query requirements. Visualization Techniques: A class of graphic techniques used to visualize the contents of a database.

SMP means (a) Systematic multi-processing (b) Stage multi-processing (c) Symmetric multi-processing (d) None of the above . (b) Removes duplication (c) Reconciles differences between various styles of data collection (d) All the above 5. Data which is not updated or changed in any way once they enter the data warehouse are called: (a) Non-volatile (b) Integrated (c) Cluster (d) Coding 4. Who is the originator of the data warehousing concept? (a) Johan McIntyre (b) Bill Inmon (c) Bill Gates (d) Chris Erickson 3. Data cleansing (a) The process of resolving inconsistencies and fixing the anomalies in source data. A data warehouse is a………. (a) Integrated and subject oriented (b) Time variant (c) Non volatile (d) All the above 2.Appendix B MULTIPLE-CHOICE QUESTIONS 1.collection of data in support of management’s decisionmaking process.

... 5 (c).. 4 (a).116 | Data Mining and Warehousing 6. Composite partitioning combines . .... 10 (a)...... 2 (b).. (a) Business houses (b) Workgroups or departmental warehouses (c) Data structures (d) Company warehouses 9. 7 (b). OLAP means (a) Off line adaptive processing (b) Off line adaptive product (c) On line adaptive product (d) On line analytical processing 10. and .. which are small in size. Data we use to run our business is (a) Historical data (b) Operational data (c) Statistical data (d) Transactional data 8. 6 (d)..... OLTP means (a) Off line transaction processing (b) Off line transaction product (c) On line transaction product (d) On line transaction processing 7.... Data marts are . 9 (d). (a) Range & hash (b) Hash & index (c) Index and range (d) None of the above Answer 1 (d). 3 (a)... partitioning... 8 (b).

but . MOLAP and HOLAP? ROLAP. ROLAP stands for Relational OLAP. Typical relational databases are designed for on-line transactional processing (OLTP) and do not meet the requirements for effective on-line analytical processing (OLAP). What is the difference between a W/H and an OLTP application? Typical relational databases are designed for on-line transactional processing (OLTP) and do not meet the requirements for effective on-line analytical processing (OLAP). For example: New MERGE command (also called UPSERT. ROLAP. External Tables allows users to run SELECT statements on external data files (with pipelining support).Appendix C FREQUENTLY ASKED QUESTIONS AND ANSWERS 1. As a result. data warehouses are designed differently than traditional relational databases. transforming (or transporting) and loading (ETL) data from source systems into the data warehouse. A data mart only contains the required subject specific data for local analysis. inquiry only collection of operational data. Many new features in the Oracle9i database will also make ETL processing easier. What is ETL/How does Oracle support the ETL process? ETL is the Data Warehouse acquisition processes of extracting. consistency and integrity (the “ACID” tests). As a result. data warehouses are designed differently than traditional relational databases. Insert and update information in one step). Since a data warehouse is not updated. these constraints are relaxed. while Data Marts is used on a business division/department level. Users see their data organized in cubes with dimensions. 4. MOLAP and HOLAP are specialized OLAP (Online Analytical Processing) applications. What is a data warehouse? A data warehouse is the “corporate memory”. What is the difference between OLAP. non-volatile (read only) and integrated. In general a Data Warehouse is used on an enterprise level. 2. Warehouses are time referenced. It is a subject oriented. point-in-time. Oracle supports the ETL process with their “Oracle Warehouse Builder” product. OLTP databases are designed to maintain atomicity. 3. What is the difference between a data warehouse and a data mart? There are inherent similarities between the basic constructs used to design a data warehouse and a data mart. subject-oriented.

with one segment for each “dimension. 5. It allows for the highest level of flexibility of metadata Low maintenance as the data warehouse matures Best possible performance. response times are usually slow. HOLAP stands for Hybrid OLAP. An ODS may contain 30 to 90 days of information. Discoverer is an ad-hoc end-user query tool. What is a star schema? Why does one design this way? Why? A single “fact table” containing a compound primary key. 6. A warehouse typically contains years of data (time referenced). The RDBMS will store data at a fine grain level. In a MOLAP system lot of queries have a finite answer and performance is usually critical and fast. 10. numeric facts. What Oracle tools can be used to design and build a W/H? Data Warehouse Builder (or Oracle data mart builder). allowing them to slice and dice the data to answer business questions. What is the difference between an ODS and a W/H? An ODS (operational data store) is an integrated database of operational data. Oracle express. Snow flake schema is similar to the star schema. Oracle designer. Express objects. 7. Other properties are: Non-volatile (read only) and Integrated. What is the difference between Oracle express and Oracle discoverer? Express is an MD database and development environment. 11. In a HOLAP system one will find queries on aggregated data as well as on detailed data. but the data is store in a Multi-dimensional database (MDBMS) like Oracle Express Server. MOLAP stands for Multidimensional OLAP. Its sources include legacy systems and it contains current or near term data. 8.” and additional columns of additive.ORA) Oracle can direct . Users see their data organized in cubes with dimensions. It can be used to represent hierarchies of information. It normalizes dimension table to save data storage space. How can Oracle materialized views be used to speed up data warehouse queries? With “Query rewrite” (QUERY_REWRITE_ENABLED=TRUE in INIT. When should one use an MD-database (multi-dimensional database) and not a relational one? Data in a multi-dimensional database is stored as business people views it. When designed correctly.118 | Data Mining and Warehousing the data is really stored in a Relational Database (RDBMS) like Oracle. Data warehouses group data by subject ather than by activity (subject-oriented). 9. an OLAP database will provide must faster response times for analytical queries. it is a combination of both worlds. Normal relational databases store data in two-dimensional tables and analytical queries against them are normally very slow. When should you use a STAR and when a SNOW-FLAKE schema? The star schema is the simplest data warehouse schema. Seagate Software’s Holos is an example HOLAP environment.

Oracle materialized views can be used to preaggregate data. 12. This constitutes two steps: Determine which dimensions will be included and determine where along the hierarchy of each dimension the information will be kept. Oracle parallel query can be used to speed up data retrieval by using multiple processes (and CPUs) to process a single task.Frequently Asked Questions & Answers | 119 queries to use pre-aggregated tables instead of scanning large tables to answer complex queries. 13. One can also use this feature to quickly move data from an OLTP database to a warehouse database. 14. What is drill down and drill up? Drill down—To expand the view to include child values that are associated with parent values in the hierarchy drill up—To collapse the list of descendant values that are associated with a parent value in the hierarchy. ETL processing using Oracle DataPump. OLAP processing using cognos. ETL processing using Data stage 7.5. Materialized views in a W/H environments is typically referred to as summaries. What Oracle features can be used to optimize the warehouse system? From Oracle8i One can transport tablespaces between Oracle databases. The query optimizer can direct queries to summary/ roll-up tables instead of the detail data tables (query rewrite). This will dramatically speed-up warehouse queries and saves valuable machine resources. . Using this feature one can easily “detach” a tablespace for archiving purposes. we mean the lowest level of information that will be stored in the fact table. The determining factors usually go back to the requirements. designing data W/H using ERWIN. exp/imp and SQL*LOADER. What tools can be used to design and build a W/H? ETL processing using informatica . 15. Data partitioning allows one to split big tables into smaller more manageable sub-tables (partitions). business objects. because they store summarized data. By granularity. Data is automatically directed to the correct partition based on data ranges or hash values. What is granularity? The first step in designing a fact table is to determine the granularity of the fact table.

Appendix D

Data mining Time: 3 hrs 5 × 15 = 75

1. a. What do you mean by expanding universe of data? b. Explain the term “data mining” (or) 2. a. Explain, what are the practical applications of data mining. b. Distinguish between data mining and query tools. 3. With a suitable diagram explain the architecture of data warehouse. (or) 4. What are the challenges in creating and maintaining a data warehouse? Explain in detail with suitable example. 5. What is OLAP? Explain the various tools available for OLAP. (or) 6. Write short notes on (i) K–nearest neighbour, (ii) Decision trees. 7. Explain the various steps involved in knowledge discovery process. (or) 8. Write short notes on (i) Data selection, (ii) Data cleaning, (iii) Data enrichment. 9. Explain about data compression. (or) 10. Explain the term denormalization.

Model Question Papers | 121

Data mining Time: 3 hrs 5 × 15 = 75

1. a. What is data mining and data warehousing? b. How computer system that can learn? c. Discuss data mining in marketing. (or) 2. a. Explain the practical applications of data mining. b. Write an account on machine learning and methodology of science. c. How query language suits to data mining operation? 3. a. Write in detail about manufacturing machine. b. Write short notes on cost justification. (or) 4. a. Give an account of the need for decision support-system for business and scientific applications. 5. Write in detail about the use of artificial neural network and genetic algorithm for data mining. (or) 6. What are OLAP tools? Explain various tools available for OLAP. 7. What are the different kinds of knowledge? How to develop a knowledge discovery database project? (or) 8. Explain the data selection, cleaning, enrichment, coding and analysis of knowledge discovery process. 9. What are the primitives of data mining? Explain the data mining primitives in SQL. (or) 10. Explain the following terminologies: Learning as compression of data set, Information content of a message, Noise and redundancy, Fuzzy database, Relational databases.

122 | Data Mining and Warehousing

Data mining Time: 3 hrs 5 × 15 = 75

1. a. What is meant by data mining? b. Compare data mining with query tools. (or) 2. a. Explain the concept of learning. b. Explain the complexity of search space with an example of a kangaroo in mist. 3. What is data warehouse? Why do we need it? Explain briefly. (or) 4. Explain the difficulties of a client/server and data warehousing. 5. Explain the knowledge discovery process in detail. (or) 6. Write a note on neural networks. 7. a. Explain the goal of KDD. b. What are the ten golden rules for reliable data mining? (or) 8. Write short notes on (i) Data selection, (ii) Data cleaning, (iii) Data enrichment. 9. Explain briefly the reconstruction foreign key relationship with suitable example. (or) 10. Explain the following: (i) Denormalization, (ii) Data mining primitives.

Data mining and warehousing Time: 3 hrs 5 × 15 = 75

1. What is meant by data mining? Compare data mining and query tools. Explain in detail about data mining in marketing. (or) 2. Explain in detail the knowledge discovery process. 3. Explain in detail about self learning computer systems and concept learning. (or)

Discuss in detail about database design schema and partitioning strategy. 15. 6. How do you test the data warehouse? Explain. 7. B. Discuss about system and data warehouse process managers.Model Question Papers | 123 4. 12. 9. enrichment and coding. (or) 8. Explain about visualization techniques and OLAP tools. 13. Explain in detail data warehouse features. Write in detail about data selection. 10. Write in detail about knowledge discover in database environment. (or) 6. 8. 5. 14. Write in detail about data mining and query tools. Write in detail about meta data and system and data warehouse process managers. 9. Explain in detail hardware physical layout security. Explain in detail hardware architecture. . 7. 2. (or) 10. Discuss in detail data mining and data warehouse. Explain about capacity planning and tuning the data warehouse. cleaning. Discuss how one can operate the data warehouse. Discuss about system process architecture design and database schema. Define data mining? Write in detail about information and production factor. Explain in detail the physical layout security. backup and recovery. Explain how you can true and test the data warehouse. 3. 11. Discuss in detail visualization techniques. Explain about service level agreement and operating the data warehouse. 4. Explain in detail self learning computer systems. Explain backup and recovery. Tech DEGREE EXAMINATIONS Data mining and warehousing Time: 3 hrs 5 × 15 = 75 1. 5.

the procedure and the need for testing the data warehouse. Explain with relevant diagrams the steps involved in designing a database schema.124 | Data Mining and Warehousing B. With a neat sketch. List the basic hardware architecture for a data warehouse. With suitable examples and relevant description justify the above statement. List the steps involved in operating the data warehouse. data mining as step in the process of knowledge discovery. b. 10. Give the necessary description also. Differentiate between data mining and data warehousing. Justify the above statement. What is market-basket analysis? Explain with an example. a. 8. With relevant examples explain how association rules can be mined from large data sets. Data mining plays a role in self learning computer systems. What is meta data? Discuss with an example. With a neat sketch explain the architecture of a data warehouse “A data warehouse contains subject oriented data”. 7. a. What are decision trees? Discuss their applications. 4. with related case studies. 2. Explain. What are the different security issues in data warehousing? 9. b. 3. 6. 5. in detail. explain the architecture of a data mining system. Tech DEGREE EXAMINATIONS Data mining and warehousing Time: 3 hrs 1. What is data warehouse tuning? Discuss the need for data warehouse tuning. 5 × 15 = 75 . Explain with neat sketch. Discuss with simple case studies the role of neural networks in data mining. With relevant examples discuss the need for backup and recovery with respect to a data warehouse.

Piatetsky-shapiro G. Prevention and Control of Nosocomial Infections.. Data Warehousing in the Real World. Epidemics: Identification and Management. Data Mining: Concepts and Techniques. Elmasri R. et al. Hamilton J. Edelstein.. Peter J. 2004..). Inman. Barquin & Herbert A. (Eds.. Margaret H.M. Dunhan. Princeton University Press.. 1971.1) Oracle Corporation. 1994. 2000. MIT Press. Smyth P. 2000. 2000.0.) (1996) Advance in Knowledge Discovery and Data Mining. Introduction to Time Series and Forecasting.B.E. Davis. Using and Managing the Data Warehouse. Building. 1997. Mozer. 1995. NY: John Wiley & Sons. M. 1994 . Addison Wesley. 1988. Brin S. Jean-mare Adamo. (1994) Time Series Analysis. 1993. Eds. Motwani R. Becker. John Wiley & Sons.W. Data Mining. Baltimore. Churchill R. Data Warehouse Design. Addison Wesley Pub. Sam Anahory Dennis Murray.BIBLIOGRAPHY BOOKS Fayyad U. 1999. Brockwell and Richard A. Tsur S. Neural Networks for Pattern Recognition. 1996. Gershenfeld). MIT Press. NY: Springer– Verlag . Pieter Adriaans Dolf Zantinge. Principles and Practice of Public Health Surveillance. Bishop. In: ACM Press. 1994. Kimball R. NY: Oxford University Press. The Data Warehouse Toolkit.. Morgan Kaufman.D. (1988) Improving the Convergence of Back Propagation Learning with Second Order Methods. and Uthurusamy K. NJ: Prentice Hall PTR. et al. Doebbeling N. Release 1 (9. NY: Oxford University Press. Kamber M. Dynamic Itemset Counting and Implications Rules for Market Basket Data. Md: Williams and Wilkins. T. A. Statistical Analysis of Time Series. Ramon C.. In: Wenzel RP (Ed). Predicting Future and Understanding the Past (Eds.. Part I and Part II. Cristopher M. Data Mining Introductory and Advanced Topics.1996. 2000. Fundamental of Database Systems. Oracle9i. 1999. NY: Springer-Verlag.C. Ullman J.. Pearson Education.. Navathe S. Englewood Cliffs.. Pearson Education. Weigend and N. Data Warehousing Guide. 1996. . MIT Press. Han J. Teutsch S. (1993) Neural Net Architecture for Temporal Sequence Processing.D. 1993. Paul Lane. Anderson. S.. (Eds. Data Mining for Association Rules and Sequential Patterns.M.

G. Alan.elsevier.. Support Decision Makers with a Data Warehouse. Artificial Neural Networks Soulie.. Vol 41(5) March 15. 1.nih. 1995. & Sorenson. Vol 33(2). 1994. November. Elaine.olapcouncil.ic. 39. Bill. 1996.html http://www-east. Artificial Neural Networks. Kloesgen. Francois (1991). The Right Server for Your Data Warehouse. Vol 1(8).org http://www. WEBSITES http://www. Aug. Khabaza. Brachman. Application Development Trends. Decision Support at Lands’ End – An Evolution. Noboru et al. IBM System t t t . No. Vol.htm http://www. Commandeering Mainframe Database for Data Warehouse Use.126 | Data Mining and Warehousing JOURNALS AND PUBLICATIONS Radding. Piatetsky-Shapiro. 605–615. 1995.. Murata.htm. Datamation Vol 41(5) March 15.nlm. Bustamente. (1991). Phyllis and http://www.html. W. Communications of the ACM. Vol. and Simoudis. http://pwp. K. G.arborsoft.J. Appleton. 11.shtm http://www. (1996) Mining Business Databases. R. T.ncbi.starnetinc.

6 G Genetic algorithm–11. 102 C CART–62 Confidence level–47 Classification–23 Cleaning–16 Clementine–3. 15. 10. 3. 32 Gain–27 Giga byte–64 . 36. 65 DB miner–61 Decision tree–24 De-duplication–17 Deep knowledge–12 Dendogram–39 Design–72. 79 Destination–99 Dimension table–75. 74 Foreign key–79 Frequent itemsets–47 D Data–66 Data cleaning–16 Data mining–2. 76 Dimensions–74 Distance–35 Domain consistency–17 DMQC–60 B Backup–101. 54 Apriori–48 Architecture–6. 99 Crossover–32 E ELDB–108 Encoding–27 Enrichment–17 Entropy–26 Exabyte–64 F Fact table–83. 19 Data selection–16 Data warehouse–5.Index A Aggregation–99 Algorithms–11 Applications–2. 20. 61 Cluster–37 Clustering–34 Coding–18 Cold backup–101 Constraints–79. 78 Association rules–46 Apriori algorithm–47 Data mart–95 Database–11.

103 Hidden knowledge–12 OLAP–12. 68 OLAP server–68 OLAP tools–5 Oracle–62. 92 Partitioning algorithm–51 Performance of–87 Petabyte–64 Prediction–23 Primary key–67 I Indexes–99 Inductive learning–8 Information–65 Interpretation–19 ID3 algorithm–28 Q K KDD–14 Keys–79 K-means algorithm–37 Knowledge–12.128 | Data Mining and Warehousing H Hash partitioning–94 Hardware–87 Hierarchical clustering–39 Hierarchies–77 Hot backup–102. 102 O P Parity–87 Partitioning–86. 65 Query–79 R Range partitioning–92 Raid–87. 36 Medicine–4 Meta data–96 Mutation–33 Mirroring–87 Multi-dimensional knowledge–12 Sampling algorithm–49 Scalability–69 Scatter diagrams–19 Schema–81 Sesi–87 Shallow knowledge–12 Snowflake schema–83 Source–97. 45. 82 SQC–88 SQL server 2000–89 Statistical analysis–11 Statistical techniques–11 Storage–68 N Neural networks–23 Neurons–24 Nonvolatile–6 Normalization–65 . 98 Spatial mining–52 Star schema–81. 88 Recovery–104 Relational databases–4 Report–109 Row–65 L Learning–8 Legacy system–10 List partitioning–93 S M Machine learning–10 Marketing–4.

19 Visualization techniques–19 View–99. 52 World wide web (www)–58 T Table–101 Tape–104 Temporal mining–52 Tuning–106 Tera byte–64 X XML–52 Xpert rule miner–62 Y Yotta byte–64 U Unsupervised learning–9 Unique–79 Z Zetta byte–64 V Visualization–11. 81 .Index | 129 Sybase–3 System–3 Support level–46 Supervized learning–8 W Web mining–51.